The Rise and Fall of Securitization

advertisement
The Rise and Fall of Securitization ∗
Sergey Chernenko
sergey.chernenko@fisher.osu.edu
The Ohio State University
Sam Hanson
shanson@hbs.edu
Harvard University
Adi Sunderam
asunderam@hbs.edu
Harvard University
May 2014
Abstract
The rise and fall of nontraditional securitizations—collateralized debt obligations and mortgagebacked securities backed by nonprime loans—played a central role in the financial crisis. Little is
known, however, about the factors that drove the pre-crisis surge in investor demand for these
products. Examining insurance companies’ and mutual funds’ holdings of fixed income
securities, we find evidence suggesting that both agency problems and neglected risks played an
important role in driving investor demand for nontraditional securitizations prior to crisis. We
also use our holdings data to shed light on the factors that drove the dramatic collapse of
securitization markets beginning in mid-2007. Contrary to conventional crisis narratives, we find
little evidence of widespread fire sales. Instead, our evidence is more consistent with the idea
that a self-amplifying buyers’ strike drove the dramatic collapse of securitization markets.
∗
We are grateful to Josh Coval, Mihir Desai, Robin Greenwood, David Scharfstein, Erik Stafford, and seminar
participants at the 2014 Fixed Income Conference, Harvard, Michigan Ross, Penn State, and the SEC/Maryland
Conference on the Regulation of Financial Markets for helpful comments. Hanson and Sunderam gratefully
acknowledge funding from the Division of Research at the Harvard Business School.
The rise and fall of nontraditional securitizations—collateralized debt obligations and
mortgage-backed securities backed by nonprime loans—was at the heart of the recent financial crisis.
Although traditional securitizations of prime mortgages, commercial mortgages, and consumer debt
had existed for decades, nontraditional securitizations grew exponentially from 2003 to mid-2007.
The primary market for nontraditional securitizations then collapsed in dramatic fashion in mid-2007,
and secondary market prices for these products fell precipitously throughout 2007 and 2008. These
price declines generated enormous losses for large financial intermediaries, ultimately imperiling
their soundness and helping to trigger a full-blown systemic crisis in 2008.
Despite general agreement on this broad narrative, surprisingly little is known about the
underlying factors that drove the pre-crisis surge in investor demand for these nontraditional
securitizations and the subsequent collapse of demand. While competing theoretical explanations
abound, there has been little detailed empirical work assessing these narratives. In this paper, we use
new micro-data on mutual funds’ and insurance companies’ holdings of fixed income securities to
shed light on the factors that drove the rise and fall of securitization.
Explanations of the rapid rise of nontraditional securitizations from 2003 to mid-2007 point to
specific investor types that drove growing demand for these products due to either “beliefs” or
“incentives.” Explanations that emphasize “beliefs” argue that some agents misunderstood the risks
of investing in securitizations. 1 For instance, demand may have been driven by rating-sensitive
investors who were misled by the AAA ratings granted to many securitizations. Alternatively,
demand may have been driven by investors who neglected the risk of a large nationwide downturn in
house prices, and thus believed that a diversified exposure to subprime mortgages was virtually
riskless.
By contrast, “incentives” explanations for the securitization boom emphasize the role of
principal-agent problems between professional investors and their principals. 2 According to these
narratives, investors took on tail risk by buying securitizations. Because these investors received
equity-like compensation, buying securitizations rationally maximized investors’ own gains while
ultimately destroying value for their principals. Agency explanations for the boom come in many
flavors, which differ primarily in where they locate the critical agency conflict within the financial
1
2
See, for instance, Coval, Jurek, and Stafford (2009), Gennaioli, Shleifer, and Vishny (2011).
See, for instance, Rajan (2005).
1
system. Some explanations emphasize agency problems between external capital providers and
managers at financial institutions. Other explanations emphasize agency conflicts between firm
management and individual firm employees (e.g., traders) who wanted to take excessive risks. By
studying both mutual funds and insurance companies, we are able to shed light on the role of both
types of incentive problems.
A key question in assessing these competing narratives is: which investors drove the boom
and bust in securitizations? For instance, in the case of “beliefs” explanations, it is natural to ask
which types of investors were most prone to these mistaken beliefs. Similarly, in the case of
“incentives” explanations of the boom, we want to know which types of investors were most prone to
these incentive problems and what factors exacerbated the incentive conflicts.
Investor heterogeneity also plays a key role in explanations of the collapse of the market for
securitizations from mid-2007 to 2009. A broad theoretical consensus points to fire sales—the forced
sale of securities by investors with high valuations to others with low valuations—as the key
amplification mechanism at work during the bust. 3 In these explanations, binding leverage constraints
forced financial intermediaries, who were natural holders of complex securitizations, to sell to other
investors with lower valuations. The resulting decline in prices further tightened intermediary
leverage constraints, leading to further forced sales. A central, yet largely untested, prediction of
these theories is that the decline in prices was associated with increased trade between levered
intermediaries and other investors. In other words, prices declined because large amounts of forced
trading significantly changed the identity of the marginal holder. An alternative to the conventional
fire-sales view is that investors staged a “buyers’ strike,” refusing to trade due to adverse selection or
some other friction. 4
In this paper, we examine data on investors’ holdings of securitizations to shed light on the
types of investors who drove the boom and bust. Specifically, we study quarterly fixed income
holdings of mutual funds and insurance companies from 2003 to 2010. These “real money”
3
See, for instance, Shleifer and Vishny (1992, 2011), Kiyotaki and Moore (1997), Lorenzoni (2008), Brunnermeier and
Pedersen (2009), Geanakopolos (2010), Stein (2012), Brunnermeier and Sanikov (2012), and Dávila (2011) for
theoretical treatments of the fire-sale amplification mechanism.
4
See, for instance, Dang, Gorton, and Holmstrom (2010), Hanson and Sunderam (2013), Milbrandt (2013), and Morris
and Shin (2012).
2
investors—intermediaries with access to stable capital from household savers—are two of the largest
investors in credit assets, owning close to 30% of all corporate bonds and private securitizations. 5
Throughout the analysis we distinguish between traditional securitizations—which were
backed by prime residential mortgages, commercial mortgages, and consumer debt—and
nontraditional securitizations that consisted of securitizations backed by nonprime home mortgages
and collateralized debt obligations. While traditional securitizations had existed for decades,
nontraditional securitizations were a far more recent innovation. And the rapid growth in
securitization markets during the 2003-2007 boom was concentrated in nontraditional products. 6
We begin by presenting two key aggregate stylized facts about the boom in nontraditional
securitizations from 2003 to 2007. First, there was significant variation in credit spreads at issuance
on highly-rated securitizations backed by different types of collateral. Spreads on AAA-rated
traditional securitizations were negligible. By contrast, spreads on AAA-rated nontraditional
securitizations were much wider, in line with spreads on BBB-rated corporate bonds. This suggests a
more nuanced view of the boom than simple tellings of either beliefs or incentives theories, which
tend to lump all AAA-rated securitizations together. Regardless of whether demand from some
investors was elevated due to incentives or beliefs, it appears that there were other investors who
clearly differentiated between traditional AAA-rated securitizations and the nontraditional AAArated structure products that were at the heart of the financial crisis.
Our second aggregate fact is that mutual funds and insurers as a whole were “below market
weight” in securitizations, particularly in nontraditional securitizations. This suggests that the
proximate surge in demand for structured finance products did not stem primarily from real-money
investors. Nonetheless, to the extent that real-money investors are affected by the same incentive
problems and expectational errors as other investors, cross-sectional patterns in their holdings can
help shed light on the drivers of the boom in securitization. Given that nontraditional securitizations
grew most explosively during the boom, we focus our analysis on these securitizations. A simple
5
According to the Flow of Funds, as of 2007Q2, total outstanding corporate and foreign bonds, which include privatelyissued securitizations, were $10.88 trillion. Of this, mutual funds (excluding money market mutual funds) owned $0.84
trillion (7%), and insurers (both life and property and casualty) owned $2.13 trillion (20%).
6
It should be noted that both traditional and nontraditional securitization markets collapsed in the bust, with traditional
securitization recovering more quickly and strongly.
3
variance decomposition underscores the role of investor heterogeneity: variation across investors, as
opposed to common variation over time, is crucial for understanding our panel of holdings data.
To flesh out the drivers of this investor-level variation, we study the relationship between a
host of investor characteristics suggested by theory and holdings of nontraditional securitizations
during the boom. For mutual funds, we focus on a key prediction of the incentives narrative:
managers facing stronger performance-flow relationship should take more risk. We find little
evidence of this. However, we find that inexperienced mutual fund managers invest significantly
more in these products than more experienced managers—despite the fact that inexperienced and
experienced managers faced similar performance-flow incentives. This is consistent with the idea that
inexperienced managers are more prone to beliefs. Moreover, we find a discrete shift away from
nontraditional securitizations for managers who were active during the market dislocations of fall
1998, consistent with theories of reinforcement learning. Finally, we find that team-managed mutual
funds invest far more in these products than individual-managed funds. This is consistent with the
social psychology literature on group polarization in risk-taking, which argues that social dynamics
can lead groups to take on beliefs that are more extreme than the beliefs of any single group
member. 7 Overall, the results are consistent with the idea that beliefs played an important role in
driving mutual fund demand for nontraditional securitizations.
Turning to insurance companies, we find that holdings of nontraditional securitizations are
concentrated among insurers that are poorly capitalized and have low financial strength ratings. In
addition, insurance companies that manage their portfolios internally hold more nontraditional
securitizations than those whose portfolios are managed externally, and insurance companies
organized as mutuals tend to hold fewer nontraditional securitizations than those organized as stock
companies. These findings are consistent with the idea that incentive conflicts play an important in
role in explaining insurer demand for nontraditional securitizations. In summary, our evidence
suggests that both beliefs and incentives are likely part of the explanation for the pre-crisis surge in
investor demand for these products.
Finally, we turn to investor behavior during the bust. We find little evidence of increased
transaction volume as prices of securitizations declined throughout late 2007 and 2008. Instead,
7
See, for example, Myers and Lamm (1976) and Isenberg (1986) in the social psychology literature and Bliss, Potter, and
Schwarz (2008) and Bär , Ciccotello , and Ruenzi (2010) in the mutual funds literature.
4
trading in nontraditional securitizations declined through the boom from 2003 to 2007 and was
extremely low during the 2007 to 2009 bust. In contrast, trading in corporate bonds remained
relatively constant from 2003 to 2010. This suggests that the collapse of the secondary market for
these products may be better described as a buyers’ strike or market freeze than a fire sale.
A brief note on our approach is in order. First, the aim of this paper is not to advance one
specific narrative of the crisis. Indeed, it is unlikely that one theory alone can completely explain a
complex phenomenon like the rise and fall of nontraditional securitization. Second, our analysis is
largely descriptive. Lacking clean natural experiments, we are not able to definitively identify causal
effects. Instead, we pursue a series of descriptive tests that can help us begin to discriminate between
alternative explanations. We carefully consider factors that may confound our interpretations of these
tests, but ultimately our tests remain descriptive. Nonetheless, we believe they are a useful first step.
In summary, our goal in this paper is to document a series of robust stylized facts that can inform
future theoretical and empirical work on the crisis. In this way, our approach is similar to that of
Griffin, Harris, Shu, and Topalugua (2011), who study investor behavior in the dot-com boom and
bust.
The plan for the paper is as follows. Section I provides some background on traditional and
nontraditional securitizations. Section II outlines different theoretical narratives of the crisis and
develops a number of testable hypotheses. Section III explains the numerous data sources we use in
the paper. Section IV analyzes our holdings data to explore drivers of the boom, while Section V uses
our data to shed light on the mechanics of the bust. Section VI offers some concluding observations.
I.
Background on Securitization
This section provides a brief background on securitization. Securitizations are created through
the process of “pooling”—assembling a pool of cash-flow-generating financial assets such as loans or
debt securities—and “tranching”—issuing claims of various priorities backed by that pool.
In the United States, the securitization of home mortgages dates back to the late 1960s and
early 1970s when various Government Sponsored Enterprises (GSEs), corporations that were
implicitly or explicitly backed by the US government, began issuing securities backed by residential
5
mortgage pools. 8 Only “conforming” mortgages—loans to borrowers with high credit (FICO) scores,
with low loan-to-value ratios, below a Congressionally-approved loan size limit (non-“Jumbo”
loans), and with sufficient documentation—are eligible to be securitized in GSE mortgage-backed
securities (GSE MBS). The market for GSE MBS grew exponentially during the 1980s, and by the
early 2000s it had overtaken the market for US Treasuries to become the single largest debt market in
the US (Fabozzi 2005).
The late 1980s saw the advent of several types of private securitizations that were not
implicitly backed by the government. We broadly refer to these securitizations as “traditional
securitizations”. 9 They include:
•
Commercial Backed Mortgage Securities (CMBS): The securitization of commercial
mortgages into CMBS was pioneered by investment banks in the early 1980s. However,
this market only grew to prominence in the early 1990s when the US Government’s
Resolution Trust Corporation began relying heavily on CMBS securitizations to sell off
the commercial real estate loan portfolios of failed depository institutions. The CMBS
market grew rapidly throughout the 1990s and 2000s.
•
Consumer Asset-Backed Securities (ABS): The securitization of credit card debt, auto
loans, and student loans began in the late 1980s. These markets developed rapidly
throughout the 1990s, and by 2000 over 30% of all non-mortgage consumer debt had been
securitized. However, the market for consumer ABS grew more slowly during the 20032007 boom, and the fraction of consumer debt that had been securitized actually fell to
25% by 2007Q2.
•
Prime non-GSE MBS: Beginning in 1977, investment banks began securitizing prime
“jumbo” home mortgages that conformed to all GSE MBS criteria other than size
(Fabozzi 2005; Gorton 2008).
8
Ginnie Mae issued its first MBS in 1968, and Freddie Mac followed in 1971. Fannie Mae did not begin issued MBS
until 1981. Ginnie Mae is a government owned corporation and is explicitly guaranteed by the US government. Fannie
Mae and Freddie Mac receive a variety of unique government benefits and have historically been perceived as being
implicitly backed by the government.
9
The exact definitions of these different types of securitizations are not universally agreed upon. We typically hew
closely to conventional investor categorizations. However, we sometimes make simplifications for the sake of clarity.
6
The boom in securitization from 2003 to 2007 prominently featured new types of
securitizations that developed after these traditional securitizations. We label the second set of
securitizations “nontraditional”. They include
•
Nonprime RMBS: The securitization of subprime and Alt-A home mortgages began in
the mid 1990s (Lehman Brothers 2004). Alt-A mortgages are non-conforming due to high
loan-to-value ratios or a lack of sufficient documentation. Subprime mortgages are nonconforming because the borrower has a low FICO score. The origination and
securitization of nonprime mortgages exploded during the boom from 2003-2007. 10
•
Collateralized Debt Obligations (CDOs): CDOs are securitizations backed by a
portfolio of fixed income assets. Depending on the underlying assets, CDOs are classified
as collateralized bond obligations (CBOs), collateralized loan obligations (CLOs), or ABS
CDOs. CBOs are collateralized largely by corporate bonds and were the most common
type of CDO in the late 1990s and early 2000s. However, this market grew slowly after
the prominent downgrades of several CBOs backed by high yield bonds in 2002. CLOs
primarily use unsecured loans to highly leveraged corporations (“levered loans”) as
collateral. CLO issuance boomed from 2003 to 2007. ABS CDOs use bonds from other
securitizations as the underlying collateral. ABS CDOs, particularly those using nonprime
RMBS as collateral, grew explosively over the course of the boom. Many of the most
dramatic losses by large financial intermediaries announced during 2007 and 2008 were
linked to ABS CDOs. 11
Although our classification of securitizations into traditional and nontraditional is not universally
used in the academic literature, the classification is meaningful and also serves as a useful short-hand.
Furthermore, it is motivated by the historical evolution of these markets described above,
classifications widely employed by practitioners, as well as by the differences in new-issue pricing
that we discuss below.
10
See Fabozzi (2005), Gorton (2008), and Ashcraft and Schuermann (2008) for further background on nonprime MBS.
Oddly, market participants sometimes refer to subprime mortgages as home equity loans (HEL). This is because the first
subprime MBS securitizations in the mid-1990s actually contained second lien mortgages or home equity loans (Lehman
Brothers 2004). Over time, these were replaced by first lien mortgages, but the name stuck.
11
See Shivdasani and Wang (2013) for further background on CLOs. See A. K. Barnett-Hart (2009) and Cordell, Huang,
and Williams (2012) for more detail on ABS CDOs.
7
II.
Narratives of the Rise and Fall
In this section, we discuss competing narratives of the boom and bust in nontraditional
securitization with an eye towards motivating our empirical tests. Our description of these narratives
and their predictions is deliberately stark. Our goal in this paper is not to provide definitive tests of
the different narratives. Instead, our goal is to document a series of robust stylized facts that can
inform future research about the mechanisms that drove the rise and fall of securitization. 12 In
pursuing this goal, our approach is to document correlations, even if they reflect endogenous
relationships between investor characteristics and holdings of structured products.
A. Narratives of the Rise
Explanations of the boom in securitization can broadly be grouped into two categories: beliefs
and incentives. 13 We describe each of these narratives in turn, outlining their empirical predictions
for prices and patterns in investor holdings.
A.1
Beliefs
In the context of the 2003-2007 securitization boom, beliefs explanations argue that investors
misunderstood the risks of investing in nontraditional securitizations. Essentially, investors treated all
AAA-rated securitizations as though they were virtually risk-free. For instance, Gerardi, Lehnert,
Sherlund, and Willen (2008) and Gennaioli, Shleifer, and Vishny (2012) argue that investors
neglected the risk of a substantial nationwide downturn in house prices and therefore believed that
diversified exposure to residential mortgages was almost riskless. Coval, Jurek, and Stafford (2010)
argue that investors focused on credit ratings which reflected expected losses, but neglected the
systematic risk embedded in securitizations. Ashcraft, Goldsmith-Pinkham, and Vickery (2010) and
Rajan, Seru, and Vig (2012) argue that investors and rating agencies failed to update their models of
12
Erel, Nadauld, and Stulz (2013) carry out a similar exercise for bank holding companies. However, they are not able to
examine variation across the type of collateral backing the securitization.
13
We do not explicitly focus on explanations of the boom that emphasize the demand for safe, “money-like” assets (e.g.,
Caballero, 2009; Caballero and Farhi, 2013; Gorton and Metrick 2009, 2010). While these explanations are surely
important, they can fall into either the beliefs or incentives categories depending on the reasons that investors perceived
nontraditional securitizations to be money-like. Similarly, we largely bypass explanations focusing on the role of
international demand for safe assets (e.g., Bernanke 2005, Bernanke, et. al. 2011, Caballero, 2009; Shin, 2012). Since our
data only included US-based mutual funds and insurers, it is not particularly useful for addressing these questions.
8
loan default in the face of deteriorating loan underwriting standards, thus neglecting the risks
associated with lending to riskier households.
We begin with predictions about prices. Simple versions of the beliefs narrative for the boom
have strong predictions for security prices. Specifically, if the demand from investors who neglected
these risks was large enough, then the yield of all AAA-rated securitizations should have been similar
and close to those on default-free government bonds.
Beliefs 1: Prices of AAA-rated nontraditional securitizations should have been highly
homogeneous and close to risk-free rates.
We next turn to predictions about holdings. In beliefs narratives, a key question is which types
of investors were the optimists who bought large quantities of nontraditional securitizations during
the boom. In the simplest version of the beliefs narrative, all investors incorrectly assessed the risks
of securitizations. This suggests that most of the variation in our holdings data should be driven by
common variation over time.
Beliefs 2: Under a simple version of the narrative, most variation in holdings of
nontraditional securitizations should be common variation over time.
In more nuanced versions of this narrative, particular subsets of investors may have been
more susceptible to beliefs. For instance, Gennaioli, Shleifer, and Vishny (2012) argue that
unsophisticated investors were shocked by the swift decline of nationwide home price. Consistent
with this, anecdotal accounts suggest that demand from smaller, less sophisticated investors like
small insurance companies and pension funds played an important role in fueling the boom. 14 Coval,
Jurek, and Stafford (2010) argue that rating-sensitive investors, and the rating agencies themselves,
were crucial because ratings measure expected loss rather than systematic risk. From a security
design perspective, the AAA-rated tranches of securitization are “economic catastrophe bonds” that
load heavily on tail risk and therefore maximize yield for a given expected loss.
In the context of mutual funds, the literature has argued that inexperienced managers may be
more prone to beliefs (e.g., Greenwood and Nagel, 2009; Malmandier and Nagel, 2011; Campbell,
Ramadorai, and Ranish, 2013). Similarly, the literature on group polarization in social psychology
14
For instance, the town of Narvik, Norway suffered significant losses on CDO investments and appeared in the infamous
stick figure “Subprime Primer” (http://www.nytimes.com/2007/12/02/world/europe/02norway.html). Springfield, MA
also suffered similar losses (http://dealbook.nytimes.com/2008/01/28/in-this-city-cdo-spells-trouble/).
9
argues that groups tend to make decisions that are more extreme than the initial inclinations of its
members. This suggests that team-managed mutual funds may take more extreme positions in
securitizations than individually-managed funds. Finally, the level of expertise of a mutual fund
family in evaluating fixed income securities may proxy for the accuracy of its beliefs. Specifically,
under the beliefs view, it seems natural to think that unsophisticated mutual funds seeking “safe”
assets would be large holders of securitizations, while more sophisticated funds would correctly
avoid these products.
Beliefs 3: Mutual funds managed by less experienced managers and mutual funds
managed by teams should have purchased more nontraditional securitizations.
A.2
Incentives
In contrast to beliefs narratives, incentives narratives of the boom emphasize agency problems
within the financial sector. 15 In these explanations of the boom, sophisticated actors within financial
institutions correctly understood the risks of securitizations all along. However, due to incentive
frictions between these agents and their ultimate principals, these agents purchased more
securitizations than their ultimate principals would have if acting on their own behalf.
For prices, the incentives narrative has similar implications to the beliefs narratives. In very
simple versions of the narrative, given enough demand from financial institutions fraught with
agency problems, the prices of AAA-rated securitizations should have been highly homogenous.
Essentially, if principals treated all AAA-rated securities as though they were risk free, agents would
demand as much of these securities as possible, regardless of their true riskiness.
Incentives 1: Under the incentives narrative, prices of AAA-rated nontraditional
securitizations should have been highly homogeneous and close to risk-free rates.
Different versions of the incentives narrative locate the critical agency friction at different
points within the financial system. Some emphasize agency problems between external capital
providers and managers at financial institutions. For instance, the mutual fund literature typically
argues that agency problems between fund managers and investors stem from the relationship
between fund performance and future flows (Chevalier and Ellison, 1997, 1998). Since expected fund
inflows are a convex function of past fund outperformance, managers face incentives to take on
15
There is a large literature studying the incentives faced by loan officers in the origination process (e.g., Keys,
Mukherjee, Seru, and Vig 2009). We focus on incentive problems with investors who ultimately purchased those loans.
10
significant risk to thus maximize their expected future compensation even if this does not maximize
expected fund returns. Thus, according to the incentives view, the stronger the performance-flow
relationship facing a fund manager, the larger her holdings of securitizations should be.
Incentives 2: Mutual funds’ holdings of nontraditional securitizations should be correlated
with the strength of the performance-flow relationship faced by funds.
Because insurance companies sometimes outsource the management of their portfolios to
external asset managers, they also offer a venue for studying this type of incentive problem. In this
kind of delegated asset management, the central conflict is between traders, who are compensated
based on risk-adjusted performance, and external capital providers. If capital providers cannot
perfectly observe risk-taking and therefore perfectly risk-adjust performance, traders may have
incentives to take excessive risks. This benchmarking problem may be particularly acute for highlyrated securitizations since, almost by construction, they perform well outside of adverse, lowprobability economic states (e.g., Acharya, Pagano, and Volpin, 2013; Adrian and Shin, 2013). 16
Incentives 3: Under the incentives narrative, insurers that have externally managed
portfolios should buy more nontraditional securitizations.
Agency problems between external capital providers and managers at financial institutions
can also take the form of risk-shifting in the tradition of Jensen and Meckling (1976). In these
narratives (e.g., Landier, Sraer, and Thesmar, 2011), managers acting on behalf of existing
shareholders bought securitizations to take on tail risk, increasing the expected payoff to equity,
while reducing total firm value at creditors’ expense. 17,18 For instance, the capitalization levels and
ratings of insurance companies may reflect the strength of the agency conflict between their equity
holders and their creditors, implicit (i.e., the government) and explicit (i.e., policyholders).19
Specifically, less well-capitalized insurers may have stronger risk-shifting incentives, and may
therefore be more likely to hold nontraditional securitizations. Finally, insurance companies are
16
Ellul and Yerramilli (2012) provide evidence that banks with stronger risk management took less tail risk in the boom.
Beltratti and Stulz (2011), Fahlenbrach and Stulz (2011), and Erel, Nadauld, and Stulz (2013) argue that incentive
conflicts between shareholders and managers have little power to explain the performance of banks during the crisis.
18
The “too big to fail” version of this story focuses on the government’s role as an implicit creditor of large financial
institutions. Under this explanation, these institutions bought securitizations to rationally maximize the value of their
implicit government guarantees. See e.g., Acharya, Cooley, Richardson, and Walter, 2009; Acharya and Richardson,
2009; Acharya, Schanbl, and Suarez, 2011; and Iannotta and Pennacchi, 2012.
19
The literature on risk-taking incentives in insurance companies is less well-developed than the comparable literature on
mutual funds. Important exceptions include Becker and Ivashina (2013), who show that within rating insurers tend to hold
riskier corporate bonds, and Koijen and Yogo (2013), who study regulatory capital arbitrage by insurers.
17
11
typically organized either as public companies with external shareholders or as mutuals, which are
owned by policyholders. These differences in organizational form also create differences in the
strength of the agency conflict between equity holders and creditors. In the case of mutual companies,
policy holders are effectively providing both equity and debt to mutual insurance companies,
potentially reducing the scope for risk-shifting.
Incentives 4: Insurers with less capital and low ratings should buy more nontraditional
securitizations. In addition, insurers organized as mutuals should buy less.
Note that both the beliefs and incentives narratives provide reasons why some investors may
have had elevated demand for nontraditional securitizations. The large literature on limits to arbitrage
initiated by DeLong et al (1990a) and Shleifer and Vishny (1997) suggests that other investors may
have been reluctant to aggressively lean against these demand shocks. This combination of demand
shocks and limited arbitrageur risk-bearing capacity would then explain why these demand shocks
raised equilibrium prices and led to a surge in issuance. Going further, rational investors may even
have had incentives to “ride” a bubble rather than leaning against it as in DeLong et al. (1990b) and
Abreu and Brunnermeier (2002).
B. Narratives of the Bust
In contrast to the strong disagreement about the drivers of the boom in securitizations, there is
wider consensus about the ensuing market collapse from mid-2007 to 2009. The frictionless null
hypothesis is that the decline in the prices of securitizations during the financial crisis was driven by
bad news about their fundamental values (i.e., news about discounted expected cash flows). If all
investors simultaneously revised their assessment of fundamental value, we would not expect the
collapse in prices to be associated with a surge in trading volume. By contrast, narratives of the bust
have typically emphasized amplification mechanisms that led prices to over-react to this bad news.
B.1
Fire sales
The most common narrative of the bust suggests that fire sales significantly amplified the
impact of this initial bad news on prices. In these explanations, binding leverage constraints forced
financial intermediaries—who were the natural holders of securitizations—to sell to other investors
with lower valuations. The resulting decline in prices further tightened leverage constraints, leading
to further forced sales. In other words, fire-sales theories articulate a natural amplification mechanism
12
whereby some initial moderately bad news could have ultimately had a severe impact on prices. 20 A
critical feature of these explanations is that the decline in the prices of securitizations was associated
with increased trade between levered intermediaries and other investors. Put differently, these
explanations for the bust emphasize fire sales not fire holds. Prices declined because the identity of
the marginal investor changed as natural holders were forced to sell their holding to others. 21
In this way, different explanations for the bust have very different implications for aggregate
trading patterns. The fundamentals-driven narrative suggests no reason for trading behavior to change
with the onset of the financial crisis, while the fire sales narrative suggests increased trade at the
beginning of the crisis.
Fire Sales 1: Under the fire sales narrative, transaction volume in nontraditional
securitizations should have risen during the crisis.
In addition, we can examine the cross-section of our transactions data to further distinguish
between different narratives of the bust. For instance, fire sales narratives predict that mutual funds
suffering large outflows are forced to liquidate their structured product holdings. A related prediction,
from the large literature on limited arbitrage, suggests that externally managed funds may have been
forced to sell more of their risky holdings than internally managed funds due to frictions between
fund managers and external capital providers.
Fire Sales 2: Under the fire sales narrative, mutual funds facing outflows and insurance
companies with externally managed portfolios should sell more nontraditional securitizations.
B.2
Buyers’ strikes
A second explanation that has received less attention is that investors’ uncertainty about the
valuations of securitizations changed dramatically. In these narratives, investors staged a “buyers’
strike”, and prices entered a downward spiral characterized by a near absence of trade. Precipitous
price declines and a collapse of trading volume may be explained by a variety of frictional
amplification mechanisms. In Diamond and Rajan (2012), for instance, institutions that would
20
See Kiyotaki and Moore, 1997; Shleifer and Vishny, 1992, 2011; Geanakoplos, 2009; Brunnermeier and Pedersen,
2009; Stein, 2012 for fire-sale explanations.
21
Merrill, Nadauld, Stulz, and Sherland (2012) provide evidence that some insurers sold some nonprime residential
mortgage backed securities at fire sale discounts. However, the aggregate volume of transactions they study is quite small.
Ellul, Jotikasthira, Lundblad, and Wang (2012) exploit a difference in regulation between property and casualty insurers
and life insurers to argue that mark-to-market accounting can exacerbate fire sales. Ellul, Jotikasthira, and Lundblad
(2011) argue that regulation forced insurers into fire sales of corporate bonds before the crisis. Coval and Stafford (2007)
study fire sales by mutual funds in equity markets.
13
otherwise sell assets at fire-sale prices instead refuse to do so due to a risk-shifting problem. By
contrast, Dang, Gorton, and Holmstrom (2010) and Hanson and Sunderam (2013) rely on adverse
selection mechanisms. In these models, bad news lowers prices and raises uncertainty about asset
valuations; as a result, the scope for adverse selection between informed and uninformed investors
becomes too large and trade collapses. Milbrandt (2012) presents a model where trade freezes
because financial intermediaries wish to avoid recognizing losses that tighten their capital constraints.
In the model, prices fall until the potential profits from transacting exceed the shadow cost of the
tightening capital constraints.
Buyers’ Strikes 1: Under the buyers’ strikes narrative, transaction volume in nontraditional
securitizations should have collapsed during the crisis.
III.
Data
The data we use in our analysis comes from a number of different sources which are
described below.
A. Mutual Fund and Insurance Holdings
Our primary data set is the Thomson Reuters eMAXX database of quarterly security-level
holdings of asset-backed securities by US-domiciled mutual funds and insurance companies.
Thomson Reuters obtains par value holdings data from regulatory filings—Schedule D for insurance
companies and Forms N-CSR(S) and N-Q for mutual funds—as well as directly from these firms.
Our sample period runs from 2003Q1 to 2010Q4, which is long enough to allow us to study changes
in the demand for securitizations during the boom as well as trading during the crisis. The sample of
firm-quarter observations conditions on having at least one asset- or mortgage-backed security in the
investment portfolio, including GSE MBS securities. Our cross-sectional results are therefore
conditional on having some exposure to securitizations. The firms that are missing from the data are
largely those that by regulation or choice never invest in securitizations. 22
22
There are two exceptions to this general statement. The first is that some money market mutual funds (MMMFs) are
included in the eMAXX data during the early part of our sample period, but disappear over time. Because of these
coverage issues and because risk taking by MMMFs has been studied before (Kacperczyk and Schnabl, 2013; Chernenko
and Sunderam, 2013), we exclude MMMFs from our analysis. The second exception is that separate accounts of
insurance companies are included in the eMAXX data during 2003-2004 and 2009Q4-2010Q4, but not during 2005Q12009Q3. We therefore exclude separate accounts from the analysis.
14
B. Securities
We supplement our holdings data by collecting detailed security-level data from a number of
sources. To obtain comprehensive data on the collateral backing each security, we collect security
data from the three major credit rating agencies—Fitch, Moody’s, and S&P—and from Bloomberg. 23
Some examples of collateral descriptions in these data are “MBS - Prime Jumbo”, “CDO - Cash ABS
- Mezzanine”, “ABS - Floorplans - Auto Dealer”, or “Subprime/Home Equity RMBS”. These
descriptions allow us to differentiate between traditional and nontraditional securitizations. To
simplify the analysis, we classify securitizations into six broad collateral types: (1) GSE MBS, (2)
CMBS, (3) consumer ABS, (4) prime private-label RMBS, (5) nonprime private-label RMBS, and (6)
CDOs. Our focus is on explaining the demand for the nontraditional securitizations that were at the
heart of the credit boom and crisis. We therefore calculate the share of nonprime RMBS and CDOs in
a given investor’s overall bond portfolio and label it nontraditional share.
In addition to information on collateral, we collect data on the coupon or spread at issuance
from Moody’s and Bloomberg, the initial credit rating from Fitch, Moody’s and S&P, and the time
series of outstanding amounts from Bloomberg.
C. Mutual Funds
We also collect data on the characteristics of the investors in our eMAXX data. We obtain
mutual fund characteristics from the CRSP Mutual Fund Database. We construct a fund-level link
between eMAXX and CRSP. We first match funds based on the spelling distance between their
names and then manually check both matched and unmatched observations. We are able to link 84%
(2,355 out of 2,813) mutual funds in eMAXX to CRSP.
Our cross-sectional analysis excludes index funds since they follow a passive investment
strategy. We also exclude funds that invest solely in government securities since the mandates of
these funds routinely restrict them from investing in private securitizations. Finally, we exclude the
handful of equity and municipal bond funds that show up in our data because they account for a very
23
eMAXX provides some information on the collateral backing each security, but the information is not granular enough
for our purposes. The most common value of the collateral variable, accounting for 68.5% of all security-level
observations, is Single Family Mortgage Loans. This is a very broad category that mixes agency securities together with
private-label subprime RMBS.
15
small fraction of aggregate securitization holdings and invest a very small share of their portfolio in
securitizations.
From Morningstar we obtain data on the portfolio managers of all US mutual funds, including
their age, educational background, and their start and end dates managing different funds. We use
birth year when available and educational background information to infer each manager’s age. We
measure manager’s experience using the first time we observe them managing a mutual fund.
D. Insurance Companies
We obtain insurance company data from AM Best. The AM Best data includes information on
insurers’ assets, capitalization, income, realized and unrealized capital gains, insurance premia, and
AM Best’s financial strength rating for individual insurers. We are able to match 84% (3,630 out of
4,302) of insurance companies in eMAXX to AM Best. Overall, our data on insurance companies is
similar to our data on mutual funds, except that we do not have biographic data on insurer portfolio
managers.
IV.
The Rise
A. Aggregate Facts
In this section, we use our data to shed light on the drivers of the boom in securitization between
2003 and 2007. As discussed in Section III, the two main explanations we explore are those based on
incentives and those based on beliefs. We start by examining aggregate time-series facts about the
boom, before turning to more detailed cross-sectional analyses.
A.1
Aggregate issuance during the boom
We begin by providing some background statistics on the explosive growth of the market for
securitizations between 2003 and 2007. Figure 1 shows the dramatic rise and fall of issuance of
securitizations. Issuance of nontraditional securitizations almost quadrupled from $98 billion in
2002Q4 to $420 billion at the peak in 2006Q4. By comparison, issuance of traditional securitizations
roughly doubled from $103 billion in 2002Q4 to $200 billion at its peak in 2007Q2. Panel B shows
that out of traditional securitizations, consumer ABS was roughly stable, and most of the growth took
place in CMBS. Panel C shows that the growth of nontraditional securitizations was dominated by
nonprime RMBS, while CDO issuance was concentrated in 2005-2006.
16
A.2
Primary market pricing during the boom
As this rapid growth in issuance and outstanding amounts was taking place, there was
surprisingly little adjustment in the prices of securitizations. In Panel A of Figure 2, we plot the
quarterly time series of average credit spreads for AAA-rated securitizations backed by different
types of collateral from 2003Q1 to 2007Q4. We compute value-weighted averages of new-issue
spreads, restricting attention to the spreads on floating rate notes indexed to 1-, 3-, and 6-month
LIBOR to avoid benchmarking issues.
Panel A shows that during the boom, credit spreads on AAA-rated nontraditional
securitizations were significantly wider than those on other comparably-rated credit assets, including
AAA-rated traditional securitizations and corporate bonds. The data reject the conventional view that
investors were willing to buy any AAA-rated asset at nearly risk-free rates. The difference in spreads
between nontraditional AAA-rated products and traditional AAA-rated bonds, while modest in
absolute terms, is economically meaningful relative to typical spreads on AAA-rated assets. Indeed,
spreads on AAA-rated nonprime RMBS, CMBS, and CDOs were closer to those on BBB-rated
corporate bonds than to those on AAA-rated corporate bonds. During the height of the boom from
2004Q1 to 2007Q2, AAA-rated consumer ABS spreads averaged 9 bps, AAA-rated nonprime RMBS
spreads averaged 22 bps, and AAA-rated CDOs spreads averaged 34 bps. This can be compared with
corporate bond spreads that averaged 4 bps (over LIBOR) for AAA-rated bonds and 37 bps for BBBrated bonds.
Panel B of Figure 2 shows similar variation in the prices of BBB-rated securitizations. BBBrated tranches of traditional consumer ABS securitizations largely traded in line with BBB-rated
corporate bonds. By contrast, BBB-rated nonprime RMBS, CMBS, and CDO tranches traded in line
with BB-rated and B-rated corporate junk bonds. Specifically, from 2004Q1 to 2007Q2, spreads on
BBB-rated consumer ABS averaged 81 bps, spreads on BBB-rated RMBS averaged 179 bps, and
those on BBB-rated CDOs averaged 231 bps. This compares with corporate bond spreads that
averaged 37 bps (over LIBOR) for BBB-rated bonds, 131 bps for BB-rated bonds, and 231 for Brated bonds.
At the same time, we find little difference in the cross-sectional dispersion of spreads within a
given rating and collateral class. For instance, from 2004Q1 to 2007Q2 the cross-sectional standard
deviation of spreads among AAA-rated consumer ABS averaged 10 bps, whereas the cross-sectional
17
dispersion for AAA-rated CDOs averaged 12 bps. Thus, the bulk of variation in credit spreads within
a given rating was across collateral classes as opposed to within collateral class. 24
The very wide spreads on nontraditional securitizations are an underappreciated fact about the
boom, and one that creates problems for simple versions of both beliefs and incentives explanations.
As discussed above, if beliefs or incentives affect the demand for all securitizations similarly, then we
would expect the prices of securitizations to be quite homogenous. For instance, taken literally, in
Gennaioli, Shleifer, and Vishny’s (2012) neglected risks model of the boom, all investors should have
treated all securitizations as a perfect substitute for riskless Treasuries. Instead, the price variation we
observe suggests that the marginal buyer of nontraditional securitizations believed these assets were
significantly more risky than traditional securitizations. 25 Moreover, the observed variation in spreads
lines up with ex post loss rates, at least directionally. Figure 3 plots average spreads on different asset
classes during the boom, against their cumulative loss rates in the bust as reported by Moody’s
(2012). The figure shows a positive correlation, indicating that the securities investors regarded as
more risky during the boom did in fact suffer more losses during the bust. This does not mean that
investors got prices “right” during the boom. Given the ex post losses, investors should have been
charging spreads an order of magnitude larger. However, it does show that investors correctly
understood the ordering of risk across products.
Of course, this does not rule out beliefs or incentives as key drivers of the boom. It merely
suggests a more complex story in which agents disagreed about the risks of nontraditional ABS, but
investors with more conservative valuations had limited risk-bearing capacity.
A.3
Who bought securitizations during the boom?
The variation in prices documented above suggests that data on investor holdings are critical
for understanding drivers of the boom. They suggest important heterogeneity in investor attitudes
towards different types of securitizations, making it important to know which intermediaries were
driving demand. Relatively little is known about which intermediaries bought subprime RMBS and
24
Thus, our results do not contradict those in Adelino (2010), who finds that, in the cross-section of AAA-rated subprime
RMBS, new issue credit spreads have very little predictive power for future downgrades or default. His findings within an
asset class and rating are consistent with our between asset class findings.
25
It seems plausible that the pricing differences between traditional and nontraditional securitizations were, to a
significant degree, about expected losses as opposed to differences in risk premia. Since all highly-rated securitization
tranches are like “economic catastrophe bonds” which only incur losses in systematically bad times, it is difficult to
appeal solely to risk premia to explain the differences between traditional and nontraditional ABS pricing.
18
CDOs during the boom. To some extent, this gap in our knowledge reflects shortcomings in
reporting: prior to the crisis, regulatory filings for banks and insurers did not readily distinguish
between securitizations and traditional corporate bonds. Fortunately, we can use our eMAXX data to
begin to fill in these holes in our understanding. 26
Tables 1 and 2 report estimates of the outstanding quantity of various private credit securities
and the estimated holdings of insurers and mutual funds from 2003 to 2010. In Table 1, we report
holdings of all private credit securities, which include corporate bonds, foreign bonds, and privately
issued ABS. In 2003Q2, insurers owned 26% of private credit securities which trended down to 20%
by 2010Q4. Mutual funds owned 7% of private credit securities in 2003Q2 which trended up to 11%
by 2010Q4. Thus, our data capture two of the largest investors in credit assets. We use these numbers
as a benchmark for assessing the overall weight of insurers and mutual funds within private credit
securities markets.
We next report holdings of traditional consumer ABS and CMBS. Both insurers and mutual
funds were underweight traditional consumer ABS. However, insurers were consistently overweight
CMBS, holding approximately one-third of outstanding CMBS from 2003 to 2010. This is consistent
with insurers’ long-standing expertise in commercial real estate. According to the Flow of Funds,
insurers’ market share of all commercial mortgages, not including CMBS holdings, averaged 28%
from 1952 to 2000. 27
Table 2 shows the evolution of outstanding amounts and holdings for nontraditional
securitizations—private RMBS and CDOs/CLOs. As shown in Table 2, outstanding RMBS and
CDOs/CLOs surged during the boom. However, both insurers and mutual funds were lightening up
on subprime RMBS, alt-A RMBS, and CDOs/CLOs exposure from 2003 to 2007. Specifically,
comparing insurers’ market shares in Panel B with their share of all private credit securities in Panel
26
Previous estimates, put together in the midst of the crisis, include Lehman Brothers (2007, 2008).
According to the National Association of Insurance Commissioners (2012), insurers invest in commercial real estate
“for three primary reasons: (1) to increase the level of asset type diversification in their investment portfolio; (2) to match
long-term assets with long-term liabilities, because commercial mortgage loans are generally long-term with fixed interest
rates and almost always include prepayment (call) protections; and (3) to minimize credit losses, because, based on
historical experience, commercial mortgage loans have had modest realized credit losses.” See
http://www.naic.org/capital_markets_archive/121026.htm for further background.
27
19
A, we see that insurers became more underweight RMBS and CDOs as issuance boomed from 2003
to 2007Q2. 28 We see similar patterns for mutual funds. 29
This suggests that demand for these assets was not primarily coming from the regulated,
unlevered sector. If regulatory arbitrage was driving the surge in investor demand, it was not the
desire of insurance companies to arbitrage their capital regulations in order to hold high-yielding
securities. Indeed, looking across collateral types, the share of real money investors appears to be
negatively correlated with the new issue credit spreads shown in Figure 2. Moreover, the distribution
of ratings for insurance company and mutual fund holdings, shown in Figures 4 and 5, shows that
they were not tilting their portfolios towards tranches with a particular rating. Approximately 80% of
their holdings are rated AAA, which is roughly the share of issuance volume that was AAA-rated.
The one exception is holdings of CDOs, where only 40% of holdings are AAA-rated. This suggests
that regulatory arbitrage was not creating demand for AAA-rated CDOs among insurers and mutual
funds.
Though the evidence suggests that mutual funds and insurers did not drive the growth of
securitizations, they were still important holders of these products, with about $90 billion in total
holdings as of 2007. In the next sections, we turn to more detailed cross-sectional determinants of
insurance company and mutual fund holdings. To the extent that these investors were affected by the
same incentives and biased beliefs as other investors, cross-sectional patterns in their holdings should
shed light on drivers of the boom in securitization more generally.
B. Mutual Funds
In this section we examine portfolio weights of nontraditional securitizations in the
investment portfolios of mutual funds, before turning to insurance companies in the next section.
B.1
Variance Decomposition
We start by reporting the results of a simple variance decomposition of the share of
nontraditional securitizations in a fund’s bond portfolio. The results in Table 3 suggest that variation
28
The National Association of Insurance Commissioners also finds that insurer holdings of CDOs are quite low. See
http://www.naic.org/capital_markets_archive/110218.htm.
29
The finding that mutual funds and insurers held very few CDOs is consistent with data on publically announced
structured finance write-downs as of January 2009. In short, with the exception of AIG, US-based life and P&C insurers
announced few CDO write-downs. Furthermore, few stand-alone commercial banks had large CDO write-downs: almost
all bank CDO write-downs were in banks with a broker-dealer. See http://www.docstoc.com/docs/45701126/Creditfluxwritedowns-Microsoft-Excel-Spreadsheet_-78-kb-xls.
20
across funds, as opposed to common variation over time, is crucial for understanding mutual fund
holdings of securitizations. Specifically, common variation over time can explain less than 2% of
total variation in the portfolio share of nontraditional securitizations. Fund fixed effects, on the other
hand, explain more than 60% of total variation.
We might expect funds within the same fund family to face similar incentives and to have
access to similar information. As a result, fund investment decisions are likely to be correlated within
a family. Columns 5-7 confirm this intuition. Fund family fixed effects can explain about 30% of
total variation in nontraditional share. Nevertheless, in large families, those with at least 10 funds,
there is significant variation across different funds that belong to the same family, suggesting that
heterogeneity across individual portfolio mandates and managers plays an important role. The R2 of
family fixed effects for these funds is only 11%.
Overall, the results of this variance decomposition underscore the role of investor
heterogeneity. To assess the beliefs and incentives narratives of the boom, we therefore examine the
cross-sectional patterns in mutual fund holdings of nontraditional securitizations.
B.2
Mutual Funds versus Variable Annuities
To begin our analysis, we focus on a key prediction of the incentives narrative: managers
facing strong performance-flow relationships should take more risk. Our ideal experiment would hold
fixed fund managers’ information and vary their incentives. This suggests comparing the investment
decisions one manager makes for two different funds with different performance-flow sensitivities.
Of course, different funds managed by the same manager are likely to have different investment
objectives and restrictions, which would confound the comparison. We would like to isolate cases
where the same manager is in charge of multiple funds with the same investment objective and
benchmark but different performance-flow sensitivities.
Variable annuities provide an excellent laboratory to approximate this ideal experiment.
Variable annuities are a type of insurance contract whose value depends on the performance of the
investment options selected by the purchaser of the insurance policy. The underlying investment
options are typically mutual funds. These variable annuity funds are regulated and structured just like
regular mutual funds under the Investment Company Act of 1940. The only difference is that variable
annuity funds are available only to purchasers of variable annuities. Because they are not widely
available and sold as part of an insurance product, variable annuity funds are likely to face different
21
performance-flow relationships, and therefore risk-taking incentives, than regular mutual funds.
Table 4 shows that variable annuity funds face lower performance-flow sensitivities than
regular mutual funds. CRSP’s coverage of variable annuities does not start until 2008Q3. 30 As a
result, we can only look at the performance-flow relationship after the crisis. Specifically, we use
monthly data for the January 20009 to June 2013 period. We include all bond and hybrid funds and
do not limit the sample to funds in our eMAXX data. Comparing the coefficients on past returns
interacted with regular and variable annuity fund dummies, performance-flow relationship is 2-3
times stronger in regular mutual funds than in variable annuity funds. Given such large difference in
the performance-flow relationship, the incentives story would predict that regular funds would
engage in more risk taking than comparable variable annuity funds.
We therefore study a sample of fund pairs where one fund is a variable annuity fund, and its
paired fund is a regular mutual fund managed by the same portfolio manager. To construct this paired
sample, we start with a sample of 328 variable annuity funds in our eMAXX data as of 2003Q4. We
then manually search for a regular mutual fund that is offered by the same fund adviser.
For example, Hartford High Yield HLS Fund is a variable annuity fund in our data. It is
offered by Hartford Life Insurance Company and its affiliates and is managed by Hartford Investment
Management Company (HIMCO). We search the list of regular mutual funds managed by HIMCO
for a regular fund with the same investment objective and find one—the Hartford High Yield Fund.
According to their prospectuses, the two funds have the same investment goal—high current income
with growth of capital a secondary objective—and investment strategy. The text describing the
investment goals and investment strategies of the two funds are virtually identical. The two funds are
also managed by the same team of three portfolio managers. The only difference between the two
funds is that Hartford High Yield HLS Fund serves as an “underlying investment option[s] for certain
variable annuity and variable life insurance separate accounts of Hartford Life Insurance Company
and its affiliates.”
We are able to find matching funds for 158 variable annuities (about one half of the original
sample). A handful of regular funds serve as matches for multiple variable annuities. Figure 6 shows
the time series of nontraditional share for matched pairs. The number of fund pairs peaks in 2003Q4,
30
CRSP does have some variable annuities before 2008Q3. But their number jumps by about an order of magnitude in
2008Q3. Furthermore, before 2008Q3, just four insurance companies account for 90% of all variable annuities.
22
the point in time as of which we actually form matched pairs, and declines over time. Nontraditional
shares track each other quite closely. Figure 6 does not control for differences in size which could in
principle affect nontraditional share. However, controlling for size does not affect our results.
Despite significant differences in the performance-flow relationship, Figure 6 shows that
managers make similar investment decisions in their regular and variable annuity funds. The fact that
we see such similar behavior despite meaningful differences in incentives points to the importance of
information and beliefs. One caveat is that in addition to facing weaker performance-flow
relationship, variable annuity funds also experience somewhat less volatile fund flows. 31 Having
more stable funding could have made variable annuity funds more willing to invest in securitizations,
offsetting the effects of facing weaker incentives. The difference in the volatility of fund flows is
smaller, however, than the difference in the performance-flow relationship. Controlling for objective
and size, the volatility of fund flows into variable annuities is lower by around 7%. This is about 13%
of the average annualized volatility of 51%.
B.3
Manager Experience
Manager experience is another characteristic that may affect the strength of the performanceflow relationship facing a mutual fund. Career concern stories emphasize the importance of strong
performance for young inexperienced managers. For instance, Chevalier and Ellison (1999) also find
that the relationship between fund flows and past performance is stronger in younger funds, and that
the termination of younger managers is more sensitive to performance than the termination of older
managers. If younger fund managers face stronger performance-flow incentives we might expect
them to tilt their portfolios towards higher-yielding, riskier securitizations. In this section, we
examine this prediction in more detail.
In our analysis, we use two alternative proxies for manager experience, both from
Morningstar. Our first proxy is age. Morningstar provides birth year information for about 20% of
managers in our data. Otherwise, we impute the age of managers based on their graduation
information (Chevalier and Ellison, 1999). Specifically, we have information on the year fund
managers received their undergraduate, masters, MBA, and PhD degrees. We assume that managers
receive these degrees when they are 22, 27, 27, and 29 years of age. These numbers correspond to the
31
Based on data for the January 2009 – June 2013 period.
23
median age of mutual fund managers receiving these degrees in our data. Our imputed age measure is
available for about 60% of fund-quarter observations in our sample.
Our second proxy for experience is the number of years each manager has been managing
mutual funds. Using the start and end dates for fund-manager pairs in our Morningstar data, we
calculate the first time each manager appears in our data. We then compute the manager’s experience
as the difference between December 31, 2004 and this date.
Of course, the experience of the fund manager is potentially endogenous to the fund’s
investment strategy. For example, a fund that intends to invest more heavily in complex
securitizations might want to hire younger portfolio managers, or more managers, to analyze these
securities. Such endogeneity concerns are one reason we measure manager experience and the
number of portfolio managers as of a single point in time. 32 To the extent that the boom in
securitization was largely unanticipated in 2004, then if we measure experience as of 2004 it is less
likely that funds chose the experience of their managers in response to the boom.
Table 5 reports summary statistics for mutual fund managers in our data. The average
manager is 45 years old and has 8.5 years of experience managing mutual funds. In addition to
experience, we also contrast the behavior of individually-managed and team-managed funds. About
64% of funds in our data are team-managed. The median fund has 2 managers.
In Figure 7 we plot the average nontraditional share of funds managed by experienced versus
inexperienced managers. The plot on the left splits managers based on their experience managing
mutual funds. Both experienced and inexperienced funds start with about a 3% portfolio weight in
nontraditional securitizations. Starting in 2004, all managers increase their nontraditional share, but
inexperienced managers increase their nontraditional share by significantly more than experienced
managers. The nontraditional share of inexperienced managers peaks at around 8% while that of
experienced managers peaks at around 5%.
The plot on the right splits managers based on their age instead of experience. Here the
picture is less clear. Although younger managers tend to have higher nontraditional shares, especially
around the peak of the credit boom, the difference is smaller at around 1%. Overall, the difference in
results highlights the importance of experience managing mutual funds as opposed to age per se.
32
Greenwood and Nagel (2009) use the same approach.
24
We examine the association between experience and nontraditional share more rigorously in
Table 6. Except for 2009, where it is close to zero, the coefficient on experience is negative and
statistically significant. Consistent with Figure 7, the coefficient is largest in magnitude during 2007.
At that point, the nontraditional share of experienced managers was 3.4% less than the nontraditional
share of inexperienced managers. This effect is economically large compared to the average
nontraditional share of 6.1%.
These splits are consistent with the idea that incentives motivated young managers to take on
more risks. However, the incentives narrative rests on the premise that young managers face stronger
performance-flow relationships. The existing literature shows that this is the case for equity mutual
funds. However, Table 7 shows it is not the case for the bond mutual funds in our data. Consistent
with prior literature, we find that fund flows respond strongly to past performance. However, when
we interact past performance with experience, we do not find any differences in the performanceflow relationship for young and inexperienced managers on one hand and old and experienced
managers on the other hand.
If not differences in incentives, what explains the tendency of young managers to invest in
nontraditional securitizations? It could be that experience directly affects manager’s beliefs. This idea
is explored by Greenwood and Nagel (2009), Malmandier and Nagel (2011a, 2011b), and Campbell,
Ramodarai, and Ranish (2013). For example, Greenwood and Nagel (2009) find that young managers
were more heavily invested in technology stocks at the peak of the technology bubble, and that they
were more likely to exhibit return-chasing behavior. In the context of bond markets, managers who
have not managed through a credit cycle, experienced sharp movements in interest rates, or sudden
unwinding of carry trades, might be less likely to fully appreciate the risks involved in investing in
securitizations.
We find evidence consistent with this in Table 8. The effect of experience seems to be highly
nonlinear—in our data it appears to matter most when a manager has at least 7-8 years of experience.
These are the managers who were active during the market dislocations of fall 1998—which saw
turbulent fluctuations in credit spreads following Russian default in August which endangered a
highly leveraged hedge fund called LTCM—and perhaps learned about risk in that episode. 33 One of
33
At the same time, it is possible that the crisis was a period during which investors learned a lot about the ability of fund
25
the key results of career concerns models is that learning about skill takes place early on. Here we
find the opposite result—little learning appears to take place for the first seven years. Overall, the
nonlinearity in the effect of experience on nontraditional share is more consistent with managers
learning about risk than with investors learning about manager ability.
B.4
Team-managed Funds
The size of the management team is another dimension along which funds may differ in their
propensity to act on biased beliefs. The social psychology literature on group polarization suggests
that social dynamics can lead groups to take on beliefs that are more extreme than the beliefs of any
single group member. For example, if the initial tendency of individual group members is to take on a
risky gamble or to engage in an altruistic act, then following group discussion, members will take on
even more extreme beliefs and attitudes.
The social psychology literature has explored two main explanations for the group
polarization phenomenon. The informational influence or persuasive argument theory suggests that a
kind of non-Bayesian updating tends to occur in groups. Essentially, group discussion tends to elicit
arguments in favor of the initially preferred alternative, and group members do not adjust for the fact
that they are exposed to a subset of information and possible arguments. 34 According to the
interpersonal comparison explanation, group members perceive the group ideal to be more extreme
than their individual positions. In an attempt to gain acceptance and to be perceived more favorably
by the group, members end up exaggerating their personal preferences.
Figure 8 shows that team-managed funds tend to have higher nontraditional share than
individually managed funds. Nontraditional share of experienced individuals is virtually flat
throughout our sample period. Inexperienced individuals appear to increase their nontraditional share
somewhat early on in 2003 and 2004, but otherwise their holdings of securitizations are mostly flat
throughout the boom. Inexperienced individuals do appear to dramatically reduce their holdings in
early 2009. In contrast, team-managed funds and, in particular, inexperienced team-managed funds
stand out. These funds more than double the share of their portfolios allocated to nontraditional
managers active at the time. According to career-concerns stories, prior learning about manager ability would result in
weaker performance-flow relationship and weaker incentives for experienced managers to invest in securitizations.
Although career concerns and learning about manager ability could potentially explain the importance of being active
during 1998, they are inconsistent with experience not mattering for managers with less than 7 years of experience.
34
This is related to the idea of “persuasion bias” explored in DeMarzo, Zwiebel, and Vayanos (2003).
26
securitizations. Once the crisis starts, their holdings fall rapidly back to the 2003-2004 levels. Table 9
formalizes these results in a regression setting. Team-managed mutual funds have 2.7% higher
nontraditional share in 2007 than individually managed funds.
These results are consistent with the idea that team-managed funds take on exposure to
nontraditional securitizations due to beliefs. Moreover, the results in Table 7 show that teammanaged funds do not face stronger performance-flow relationships. Thus, differences in incentives
are unlikely to explain our results.
Overall, our analysis of the cross section of mutual fund holdings of securitizations is more
consistent with beliefs narratives of the boom than incentives narratives. The securitizations holdings
of mutual funds do not appear to vary with the strength of the performance-flow relationships they
face. However, funds managed by teams and young individuals take more risk, consistent with beliefs
narratives. Of course, this evidence is simply suggestive, not definitive. Moreover, mutual funds may
not face the type of incentive problems relevant for incentives narratives of the boom. Therefore, we
next turn to our data on insurance company holdings.
C. Insurance Companies
Insurance companies are the single largest institutional holder of credit securities. Moreover,
they are large regulated financial intermediaries, so their holdings may give us insight into the
behavior of other such intermediaries including banks and broker dealers.
C.1
Variance Decomposition
We again start by a simple variance decomposition of the share of nontraditional
securitizations in an insurers’ portfolio. As with mutual funds above, the results in Table 10 suggest
that variation across insurers, as opposed to common variation over time, is crucial for understanding
insurer holdings of securitizations. Specifically, common variation over time can explain less than
1% of total variation in the portfolio share of nontraditional securitizations. Insurer fixed effects, on
the other hand, explain more than 58% of total variation.
Among insurers, life insurers account for roughly 18% of corporate bond holdings, while
property and casualty (P&C) insurers account for 3%. Figure 9a shows that P&C insurers tend to
invest less in nontraditional securitizations than life insurers, though the time-series patterns are
similar for both types. Holdings fall in 2003, rise from 2004-2008, and then fall afterwards. The
difference in levels between P&C and life insurance holdings of nontraditional securitizations is in
27
line with the general tendency of P&C insurers to have safer portfolios. For instance, according to the
2007Q2 Flow of Funds, Treasuries and Agencies made up 17% of the total assets of P&C insurers. In
contrast, Treasuries and Agencies made up 10% of total life insurance assets. These differences can
be explained by differences in regulations and the demand for liquidity of P&C and life insurers.
Specifically, P&C insurers have a stronger precautionary demand for liquidity because the payouts on
their policies tend to be lumpier and harder to predict than the payouts on life policies. Indeed, as we
will see in Section V.B, P&C insurers tend to transact more than life insurers, particularly in their
holdings of GSE MBS.
C.2
Financial Strength
We now turn to more detailed analysis of cross-sectional determinants of insurance company
holdings. Insurance companies offer an interesting laboratory for testing different versions of the
incentives view. There are a number of frictions that could generate agency problems within
insurance companies. For instance, equity holders are residual claimants on the values of insurers’
asset portfolios net of policyholder claims. Thus, insurers’ capitalization levels and financial strength
ratings may reflect the strength of the agency conflict between their equity holders and their creditors.
To combat this problem, regulation requires insurance companies to maintain minimum levels of
capital on a risk-adjusted basis. That is, like banks, insurers with riskier portfolios are subject to
higher capital requirements. However, to the extent that these regulations are imperfect, less wellcapitalized insurers may have stronger risk-shifting incentives, and therefore be more likely to hold
nontraditional securitizations.
To test this hypothesis, in Figure 9c we split insurers by their capital-assets ratios as of 2003.
Since capital-assets ratios are correlated with size, we define high-capital insurers to be those with
above median capital-assets ratios within a given size decile (based on total assets). We hold the
classification constant over time to ensure that the results are not driven by a compositional shift in
the set of insurers that have low versus high capital ratios.
Figure 9c shows insurers with low and high capital ratios initially have similar holdings of
nontraditional securitizations in 2003. Over the course of the boom, both types of insurers increase
their portfolio allocations to securitizations. However, insurers with low capital ratios increase their
portfolio weights far more than insurers with high capital ratios. This is consistent with the notion
that these poorly capitalized insurers engaged in risk-shifting. Tables 11 and 12 formalize this
28
evidence in regressions. Table 11 shows that the level of nontraditional share is negatively correlated
with insurer capital ratios, with the magnitude of the effects growing over time. The effect is
economically significant. The standard deviation of the capital-assets ratio is around 20%, so that a
one standard deviation increase in capitalization is associated with a 2.4% (percentage points) lower
nontraditional share in 2007, relative to a mean nontraditional share of 2.2%. Panel B of Table 11
shows that the results also hold in differences. Between 2003 and 2007, insurers with low capital
ratios increase their nontraditional share more than insurers with high capital ratios.
Of course, the capital-assets ratio is a crude proxy for insolvency risk, which is the ultimate
driver of risk-shifting behavior. For instance, insurers may have a similar risk of insolvency, but
optimally have different capital because their policy liabilities have different risks. To address this
concern, we obtain data from A.M. Best on the financial strength of insurers. A.M. Best is a credit
rating agency dedicated to the insurance industry. Its credit ratings are based on a quantitative and
qualitative evaluation of an insurer’s ability to meet its ongoing policy and contract obligations.
In Figure 9b, we split insurers into those with high and low financial strength ratings as of
2003. To ensure that the results are not driven by size, we again define highly-rated insurers to be
those with above median ratings within a given size decile. The figure shows a pattern similar to our
splits based on capital ratios. Low-rated insurers have initial portfolio allocations to nontraditional
securitizations similar to those of high-rated insurers. However, over the course of the boom, lowrated insurers increase their nontraditional shares more than high-rated insurers. Table 12 formalizes
this evidence in a regression setting. Again, the magnitudes are economically and statistically
significant. In 2007, highly-rated insurers have nontraditional shares 2.3 percentage points lower than
low-rated insurers, relative to a mean nontraditional share of 2.4%. Overall, although not conclusive,
the evidence is consistent with the notion that financially weaker insurance companies engaged in
risk-shifting by purchasing nontraditional securitizations.
C.3
Organizational Form
Insurers are organized as either stock companies with external shareholders or mutuals which
are owned by policyholders. These differences in organizational form may create differences in the
strength of the agency conflict between equity holders and other claimants. Specifically, in the case
of mutual companies, policy holders effectively provide both equity and debt, which may reduce the
scope for risk-shifting.
29
To examine this difference, Figure 9d compares the nontraditional shares of mutual insurance
companies versus non-mutuals. Note that here we cannot adjust for size in the way we did when
examining insurer capitalization, since insurers are either mutuals or not. Nevertheless, Figure 9d
shows that mutual insurance companies tend to have lower nontraditional shares than non-mutuals
and this difference increases over the 2003-2007 boom in securitization. Table 13 shows that this
difference is not statistically significant. However, the magnitudes are still economically meaningful.
Mutuals have about a nontraditional share about 1% lower than non-mutuals of the same size,
consistent with the notion that agency conflicts between external shareholders and insurance
company policyholders could affect holdings of nontraditional securitizations.
In addition to variation in their organizational form, insurers also vary in the way they manage
their investment portfolios. Specifically, insurance companies sometimes outsource the management
of their portfolios to external managers. The difference between internally managed portfolios and
externally managed portfolios may shed light on the impact of agency frictions on portfolio
allocations to nontraditional securitizations. On one hand, external managers may have incentives to
take excessive risk due to performance-based compensation. On the other hand, internal risk
management may bring its own set of agency problems.
To examine this difference, Figure 9f compares the nontraditional shares of internally
managed versus externally managed insurance portfolios. Again, we cannot adjust for size here since
portfolios are either internally managed or not. Nevertheless, Figure 9f shows that internally managed
portfolios tend to have higher nontraditional shares than externally managed portfolios during the
2003-2007 boom in securitizations. This is consistent with the idea that externally managed portfolios
are more subject to stricter limits on portfolio allocations, possibly due to agency conflicts. Table 14
shows that the difference in levels is not statistically significant. However, the magnitudes of the
point estimates are still economically meaningful. Externally managed portfolios have about a
nontraditional share about 1% lower than internally managed portfolios for insurers of the same size.
In summary, our analysis of insurance companies suggests that incentive problems may have
played an important role in driving their demand for nontraditional securitizations.
V.
The Fall
Overall, our evidence suggests that both beliefs and incentives played important roles in
30
driving the boom in securitization. We now turn to the collapse of the market for securitizations from
2007 to 2009.
A. Aggregate Evidence
As discussed in the hypothesis development section, the conventional macro-finance view of
the collapse is that there were significant fire sales of securitizations by leveraged intermediaries
during this period. This view suggests that we should observe high transaction volume during the
crisis, as the leveraged sector sells securitizations to the unlevered investors in our data. In contrast,
narratives of a “buyers’ strike” or market freeze suggest there should be little transaction volume
during the crisis.
Figure 10 shows the time series of insurer transactions by collateral type. We plot turnover,
the absolute value of transactions (i.e. buys + sells) normalized by total holdings. Thus, if an insurer
sells its entire existing portfolio and buys a portfolio of new securities, turnover will be 200%. We
take a four-quarter moving average to smooth the data series. As reference points, we also plot
turnover for GSE MBS, which are in our eMAXX data, and corporate bonds. 35
The figure shows that, during the boom, securitizations generally traded less than GSE MBS
and more than corporate bonds. Turnover in nonprime RMBS and traditional consumer ABS is close
to turnover in GSE MBS pre-crisis, while CMBS and CDOs trade substantially less. As the crisis sets
in, trade in all types of structured falls substantially. In contrast, turnover of GSE MBS and corporate
bonds remains relatively constant throughout the crisis. Overall, there is little evidence that real
money investors bought large amounts of securitizations in the crisis. There was little trade, and
overall par dollars of these products held by mutual fund and insurers fall during the crisis. 36
Figure 11 shows similar patterns for mutual funds. For mutual funds, we do not directly
observe transactions; we only observe changes in holdings from one quarter to the next. We therefore
impute transactions by adjusting the change in the holdings of each security for the change in the
security’s overall outstanding amount due to defaults and prepayments. We scale sales and purchases
35
We obtain data on insurance company corporate bond transactions from Mergent FISD. The underlying source of the
data is the same set of regulatory filings that eMAXX collects our securitizations transaction data from.
36
The fall in par dollars held does not reflect trade. It instead reflects prepayments and defaults. When the collateral in a
securitization prepays or defaults, the (par) dollar holdings of an investor in the securitization naturally fall. The crisis was
associated with both defaults and mortgage prepayments. We use data from Bloomberg on total CUSIP-level amounts
outstanding to verify that the decline in holdings is driven by prepayments rather than trade. Insurers and mutual funds
hold a constant fraction of the amount outstanding, but the total outstanding declines due to defaults and prepayments.
31
by average holdings during the previous and current quarters, and then take a four-quarter moving
average to smooth the data series. The figure shows that the turnover of GSE MBS is about twice as
high as the turnover of non-agency asset-backed securities. The turnover of nonprime RMBS trends
down over time from around 35% in late 2004 to around 10% by the end of the sample period. The
decline in nonprime RMBS turnover is especially pronounced during the crisis: it falls from 27% in
2007Q2 to 14% in 2008Q4. The turnover of CMBS and CDOs is relatively stable before the start of
the crisis in the second half of 2007 and experiences similarly large declines during the crisis. Finally,
the turnover of GSE MBS and traditional consumer ABS picks up during the crisis, potentially
consistent with funds choosing to transact in these more liquid securities.
These aggregate facts are inconsistent with simple fire sales stories like Kyotaki and Moore
(1997), Geanakoplos (2009), Brunnermeier and Pedersen (2009), Shleifer and Vishny (1992), and
Stein (2012). In most fire-sales narratives, outsized declines in prices are mediated by high trade
volume. Unless the demand curve for securitizations was very inelastic, the tiny volume of secondary
trade we observe seems insufficient to significantly depress equilibrium prices. 37 Overall, given that
both prices and trading volume drop substantially, the buyers’ strike narrative seems to be more
consistent with the data.
One issue that arises with our analysis is that we do not observe the full universe of holders or
transactions. Thus, we can only make direct statements about transactions where at least one party is
either an insurance company or mutual fund. However, as discussed above, our data covers an
economically significant portion of the universe of securitizations, and insurers and mutual funds
make up a large fraction of total credit investor assets. One possibility is there were wide-spread firesales that led to a reallocation of securitizations within the class of levered intermediaries, but not
between leveraged and unlevered intermediaries. However, to the extent that fire sales generally
result in the transfer of security holdings from one broad class of financial intermediary (levered
investors) to another (unlevered investors), it is natural to think that our data should capture at least
one side of the transaction.
37
For instance, one way to generate an extremely inelastic demand curve in a fire-sale setting would be to posit two types
of investors: natural buyers with high valuations and unnatural buyers with very low valuations. Suppose the supply of
the risky asset is 100 units, all initially held by natural buyers with high valuations. Now, if some shock forces natural
buyers to sell even 1 unit of their holdings, unnatural buyers with low valuations will become the marginal holders and
price will plummet precipitously. Of course, this is a knife-edge result. It relies on the risk-bearing capacity of natural
buyers being just enough to hold the entire asset supply before the shock.
32
Overall, the evidence appears most consistent with the notion that the collapse of the
secondary market for securitizations was driven by a buyer’s strike rather than a fire sale. In models
of buyers’ strikes, price declines spiral without any trade. While collapse happens instantly in the
models discussed in Section II, one can imagine the mechanisms spiraling more slowly over the
course of the crisis in practice. For instance, in adverse selection driven models, increases in the
uncertainty of uninformed investors might dampen trading, in turn causing uncertainty to rise further.
B. Cross-sectional Evidence
We next turn to cross-sectional analyses of our transactions data. While the aggregate data
show that the total volume of transactions was small during the bust, there may still be cross-sectional
evidence that certain insurers or mutual funds were forced to sell during bust. 38
We begin by analyzing insurance company transactions in Table 15. The table shows that
larger insurers tended to buy more nontraditional securitizations during the boom, consistent with
their higher holdings of those products. There is some evidence that larger insurers also tended to
increase their purchases in the crisis, but the magnitude of the effect is quite small. Otherwise, there
is little evidence that insurers were forced to sell nontraditional securitizations during the bust. If
anything, there is evidence that stronger insurers with higher financial strength ratings sold more.
These are unlikely to be forced sales. Insurers appear to be much more active in transacting in the
GSE MBS market. Large insurers are net sellers of GSE MBS in the boom and become increasing net
sellers during the crisis.
We next turn to mutual fund transactions in Table 16. The table shows that mutual funds
experiencing inflows tend to make net purchases of nontraditional securitizations during the boom, a
trend that does not change in the crisis. However, mutual funds experiencing outflows during the
boom do not tend to sell their nontraditional securitizations. This trend continues in the crisis—there
is no evidence that mutual funds suffering outflows are forced to sell their nontraditional
securitizations. Mutual funds with high nontraditional share of course purchased more nontraditional
securitizations during the boom. These funds ceased their purchases in the bust, but there is no
evidence that they actively sold off their holdings during the crisis.
Like insurers, mutual funds appear to be much more active transacting in the GSE MBS
38
Indeed, Merrill, Nadauld, Stulz, and Sherland (2012) provide evidence that P&C insurers sold some nonprime
residential mortgage backed securities at fire sale discounts during the crisis.
33
market. The sensitivity of GSE MBS purchases to inflows and outflows is much higher than the
sensitivity of nontraditional securitization purchases. When mutual funds have inflows, they tend to
increase their GSE MBS purchases. When they suffer outflows, they tend to sell GSE MBS.
Moreover, these sensitivities increase in the crisis.
Overall, the results are consistent with the Scholes (2000) model of liquidity management.
Mutual funds and insurers that faced liquidity needs transacted in the most liquid securitization
market, the market for GSE MBS. If there were forced sales in any secondary markets for
securitizations, it seems likely that those forced sales would take place in the market for GSE MBS.
Indeed, some have argued that the Federal Reserve’s first quantitative easing program was so
effective because it provided liquidity in the market for GSE MBS at a time when there was high
demand for liquidity in that market. 39 Our results are consistent with this notion.
VI.
Conclusion
Nontraditional securitizations—nonprime RMBS and CDOs—were at the heart of the recent
financial crisis. The demand for these securities fed the housing boom during the early and mid2000s, while declines in their prices during 2007-2008 generated large losses for financial
intermediaries, ultimately imperiling their soundness and triggering a full-blown crisis. We have
many anecdotal narratives and theoretical models of the rise and fall of nontraditional securitizations,
but little systematic empirical evidence. Using novel micro-data on insurance companies’ and mutual
funds’ holdings of both traditional and nontraditional securitizations, this paper begins to shed light
on the factors that drove the rise and fall of securitization.
We document a number of robust stylized facts that should inform future theoretical and
empirical work on the crisis. First, insurance companies and mutual funds as a whole were below
market weight in securitizations, in particular nontraditional securitizations. In 2007Q2, right before
the start of the crisis, insurance companies and mutual funds held 27.3% of all private credit
securities but only 4.3% of nontraditional securitizations. This suggests that the proximate surge in
demand for structured products did not stem from these real-money investors.
39
See Sack (2009), who argues that one reason the Federal Reserve’s first quantitative easing program had a “large
impact on MBS rates, is that the market began from a point of substantial spreads—ones that were well above market
norms. These wide spreads could have reflected poor liquidity and an elevated liquidity premium on these securities.”
Krishnamurthy and Vissing-Jorgensen (2013) make a similar argument about the first round of Fed asset purchases.
34
Second, heterogeneity across security classes and investors is key to understanding the crisis.
While spreads on AAA-rated traditional securitizations were negligible, spreads on AAA-rated
nontraditional securitizations were much wider, in line with spreads on BBB-rated corporate bonds.
This suggests a more nuanced view of the boom than the simple tellings of either beliefs or incentives
theories, which tend to lump all AAA-rated securitizations together. There were some investors who
understood and differentiated between the AAA tranches of traditional and nontraditional
securitizations. Simple variance decomposition underlines the importance of investor heterogeneity:
variation across investors as opposed to common variation over time is crucial for understanding our
holdings data.
The behavior of mutual funds is more consistent with beliefs than with incentives. We find
little evidence that funds facing stronger performance-flow relationship took on more risk by
investing in nontraditional securitizations. Despite not facing stronger performance-flow relationship,
inexperienced mutual fund managers invested significantly more in these products than did more
experienced managers. This is consistent with the idea that inexperienced managers are more prone to
beliefs. Supporting this idea is a discrete shift away from nontraditional securitizations by managers
who were active during the market dislocations of fall 1998.
Turning to insurance companies, we find that holdings of nontraditional securitizations are
concentrated among insurers that are poorly capitalized and have low financial strength ratings. In
addition, insurance companies that manage their portfolios internally hold more nontraditional
securitizations than those whose portfolios are managed externally, and insurance companies
organized as mutuals tend to hold fewer nontraditional securitizations than those organized as stock
companies. These findings are consistent with the idea that incentive conflicts play an important in
role in explaining insurer holdings of nontraditional securitizations. In summary, our evidence
suggests that both beliefs and incentives are likely part of the explanation for the pre-crisis surge in
investor demand for these products.
In contrast to multiple narratives of the boom, explanations of the bust center on a single
narrative of fire sales—the forced sales of securities by investors with high valuations to investors
with low valuations. A central, yet largely untested, prediction of fire sales models is that the decline
in prices was associated with increased trade between levered intermediaries and other investors. We
find no evidence of increased transaction volume as prices of securitizations declined throughout late
35
2007 and 2008. Instead, turnover of nontraditional securitizations experienced large declines during
the crisis. Turnover of GSE MBS, and to a lesser extent of traditional consumer ABS, increased; and
to the extent that mutual funds suffered outflows, they met redemptions through sales of GSE MBS.
Our results therefore challenge the importance of fire sales during the crisis, or at least the canonical
models of fire sales, and suggest that the collapse of the secondary market may be better described as
a buyers’ strike.
We hope that future research builds on the stylized facts documented in our paper to shed
further light on the rise and fall of securitization.
36
References
Acharya, Viral V., Thomas Cooley, Matthew Richardson, and Ingo Walter, 2010, Manufacturing Tail
Risk: A Perspective on the Financial Crisis of 2007-09, Foundations and Trends in Finance 4, 247325.
Acharya, Viral V., Thomas Cooley, Matthew Richardson, and Ingo Walter, 2010, Market Failures
and Regulatory Failures: Lessons from Past and Present Crises, Masahiro Kawai and Eswar Prasad,
ed.: Financial Sector Regulation and Reforms in Emerging Markets (Brookings Institution Press).
Acharya, Viral V., Marco Pagano, and Paolo Volpin, 2013, Seeking Alpha: Excess Risk Taking and
Competition for Managerial Talent, NBER Working Paper 18891.
Acharya, Viral V, Philipp Schnabl, and Gustavo A Suarez, 2012, Securitization without risk transfer,
Journal of Financial Economics 107, 515-536.
Adrian, Tobias, and Hyun Song Shin, 2013, Procyclical Leverage and Value at Risk, Review of
Financial Studies, forthcoming.
Ashcraft, Adam B., and Til Schuermann, 2008, Understanding the Securitization of Subprime
Mortgage Credit, Foundations and Trends in Finance 2, 191-309.
Ashcraft, Adam B., Paul Goldsmith-Pinkham, and James Vickery, 2010, MBS ratings and the
mortgage credit boom, Federal Reserve Bank of New York Staff Report 449.
Bär, Michaela, Conrad S. Ciccotello, and Stefan Ruenzi, 2010, Risk management and team-managed
mutual funds, Journal of Risk Management in Financial Institutions 4, 57–73.
Barnett-Hart, Anna Katherine, 2009, The Story of the CDO Market Meltdown: An Empirical
Analysis, Harvard College Undergraduate Thesis.
Becker, Bo, and Victoria Ivashina, 2012, Reaching for Yield in the Bond Market, Working Paper.
Beltratti, Andrea, and Rene M. Stulz, 2012, The credit crisis around the globe: Why did some banks
perform better? Journal of Financial Economics 105, 1–17.
Bernanke, Ben S., 2005, Monetary Policy in a World of Mobile Capital, Cato Journal 25, 1-12.
Bernanke, Ben S., Carol Bertaut, Laurie Pounder DeMarco, and Steven Kamin, 2011, International
Capital Flows and the Returns to Safe Assets in the United States, 2003-2007, Board of Governors of
the Federal Reserve System International Finance Discussion Paper 1014.
Black, Fisher, 1986, Noise, The Journal of Finance 41, 529–543.
Bliss, Richard T, Mark E Potter, and Christopher Schwarz, 2008, Performance Characteristics of
Individually-Managed versus Team-Managed Mutual Funds, Journal of Portfolio Management 34,
110-119.
37
Brunnermeier, Markus, and Yuliy Sannikov, 2013, A Macroeconomic Model with a Financial Sector,
American Economic Review, in press.
Brunnermeier, Markus, and Lasse Pedersen, 2009, Market Liquidity and Funding Liquidity, Review
of Financial Studies 22, 2201–2238.
Caballero, Ricardo, 2010, The “Other” Imbalance and the Financial Crisis, NBER Working Paper
15636.
Caballero, Ricardo and Emmanuel Farhi, 2013, A Model of the Safe Asset Mechanism (SAM):
Safety Traps and Economic Policy, Working Paper.
Chernenko, Sergey, and Aditya Vikram Sunderam, 2013, Frictions in Shadow Banking: Evidence
from the Lending Behavior of Money Market Mutual Funds, Working Paper.
Chevalier, Judith, and Glenn Ellison, 1997, Risk Taking by Mutual Funds as a Response to
Incentives, Journal of Political Economy 105, 1167–1200.
Chevalier, Judith, and Glenn Ellison, 1999, Career concerns of mutual fund managers, The Quarterly
Journal of Economics 114, 389–432.
Cordell, Larry, Yilin Huang, and Meredith Williams, 2012, Collateral damage: Sizing and assessing
the subprime CDO crisis, Federal Reserve Bank of Philadelphia Working Paper 11-30.
Coval, Joshua, Standard & Poor's, 2007, Asset fire sales (and purchases) in equity markets, Journal
of Financial Economics 86, 479–512.
Coval, Joshua, Jakub Jurek, and Erik Stafford, 2009, The Economics of Structured Finance, Journal
of Economic Perspectives 23: 3-25.
Dang, Tri Vi, Gary B. Gorton, and Bengt Holmstrom, 2011, Ignorance, Debt and Financial Crises,
Working Paper.
Dávila, Eduardo, 2011, Dissecting Fire Sales Externalities, Working Paper.
De Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann, 1990, Noise
Trader Risk in Financial Markets, The Journal of Political Economy 98, 703-738.
Ellul, Andrew, and Vijay Yerramilli, 2012, Stronger Risk Controls, Lower Risk: Evidence from U.S.
Bank Holding Companies, Journal of Finance, Forthcoming.
Erel, Isil, Taylor Nadauld, Rene M. Stulz, 2013, Why Did U.S. Banks Invest in Highly-Rated
Securitization Tranches? Working Paper.
Fabozzi, Frank J., 2005, The Structured Finance Market: An Investor’s Perspective, Financial
Analysts Journal, 27-40.
38
Fahlenbrach, Rüdiger, and René M. Stulz, 2011, Bank CEO Incentives and the Credit Crisis, Journal
of Financial Economics 99, 11-26.
Geanakopolos, John, 2009, The Leverage Cycle, in D. Acemoglu, K. Rogoff, and M. Woodford, eds.:
NBER Macroeconomics Annual 2009, pp. 1-65 (University of Chicago Press).
Geanakopolos, John, 2010, Incorporating Financial Features into Macroeconomics: Discussion, in
Macroeconomic Challenges: The Decade Ahead. Jackson Hole, Federal Reserve Bank of Kansas
City Economic Policy Symposium, 2010.
Gennaioli, Nicola, Andrei Shleifer, and Robert Vishny, 2012, Journal of Financial Economics,
Journal of Financial Economics 104, 452–468.
Gerardi, Kristopher, Andreas Lehnert, Shane M. Sherlund, and Paul Willen, 2008, Making Sense of
the Subprime Crisis, Brookings Papers on Economic Activity 39(2), 69-159.
Gorton, Gary B., 2008, The Panic of 2007, in Maintaining Stability in a Changing Financial System,
Proceedings of the 2008 Jackson Hole Conference, Federal Reserve Bank of Kansas City, 2008.
Gorton, Gary B., and Andrew Metrick, 2010, Regulating the Shadow Banking System, Brookings
Papers on Economic Activity 2010, 261-312.
Gorton, Gary B., and Andrew Metrick, Haircuts, Federal Reserve Bank of St. Louis Review 92(6).
Greenwood, Robin, and Stefan Nagel, 2009, Inexperienced investors and bubbles, Journal of
Financial Economics 93, 239–258.
Hanson, Samuel Gregory, and Aditya Vikram Sunderam, 2013, Are There Too Many Safe
Securities? Securitization and the Incentives for Information Production, Journal of Financial
Economics 108, 565–584.
Harrison, J Michael, and David M. Kreps, 1978, Speculative Investor Behavior in a Stock Market
with Heterogeneous Expectations, The Quarterly Journal of Economics 92(2), 323-36.
Hong, Harrison, and Jeremy C. Stein, 2003, Differences of Opinion, Short-Sales Constraints and
Market Crashes, Review of Financial Studies 16, 487-525.
Iannotta, Giuliano, and George Pennacchi, 2012, Bank Regulation, Credit Ratings, and Systematic
Risk, Working Paper.
Isenberg, Daniel J., 1986, Group polarization: A critical review and meta-analysis, Journal of
Personal and Social Psychology 50(6), 1141-1151.
Jensen, Michael, and William H. Meckling, 1976, Theory of the Firm: Managerial Behavior, Agency
Costs and Ownership Structure, Journal of Financial Economics 3, 305-360.
Kacperczyk, Marcin, and Philipp Schnabl, 2013, How safe are money market funds? Quarterly
Journal of Economics 128, 1073–1122.
39
Keys, Benjamin J., Tanmoy K. Mukherjee, Amit Seru, and Vikrant Vig, 2009, Did Securitization
Lead to Lax Screening? Evidence from Subprime Loans, Quarterly Journal of Economics 125, 307362.
Kiyotaki, Nobuhiro, and John Moore, 1997, Credit Cycles, Journal of Political Economy 105, 211248.
Landier, Augustin, David Sraer, and David Thesmar, 2011. The Risk-Shifting Hypothesis, IDEI
Working Paper 699.
Lorenzoni, G., 2008, Inefficient Credit Booms, Review of Economic Studies 75, 809-833.
Merrill, Craig, Taylor Nadauld, Rene M Stulz, and Shane M Sherlund, 2013, Why did financial
institutions sell RMBS at fire sale prices during the financial crisis? Charles A. Dice Working Paper
2013-2.
Milbradt, Konstantin, 2012, Level 3 Assets: Booking Profits, Concealing Losses, Review of Financial
Studies 25, 55-95.
Miller, Edward M., 1977, Risk, Uncertainty, and Divergence of Opinion, The Journal of Finance 32,
1151-1168.
Myers, David G, and Helmut Lamm, 1976, The Group Polarization Phenomenon, Psychological
Bulletin 83, 602–627.
Morris, Stephen, and Hyun Song Shin, 2012, Contagious Adverse Selection, American Economic
Journal: Macroeconomics 4, 1-21.
Rajan, Uday, Amir Seru, and Vikrant Vig, 2012, The Failure of Models that Predict Failure:
Distance, Incentives and Defaults, Journal of Financial Economics, accepted.
Scheinkman, Jose, and Wei Xiong, 2003, Overconfidence and speculative bubbles, Journal of
Political Economy 111, 1183–1219.
Shin, H. “Global Banking Glut and Loan Risk Premium”, Mundell-Fleming Lecture, IMF Economic
Review 60 (2), 155-192, 2012
Shleifer, Andrei, and Robert Vishny, 1992, Liquidation Values and Debt Capacity: A Market
Equilibrium Approach, Journal of Finance 47, 1343-1366.
Shleifer, Andrei and Robert Vishny, 2011, Fire Sales in Finance and Macroeconomics, Journal of
Economic Perspectives 25, 29-48.
40
Figure 1
Issuance of Traditional and Nontraditional Securitizations
This figure shows quarterly issuance of traditional and nontraditional securitizations in SDC data.
Traditional securitizations include CMBS, prime RMBS, consumer ABS, and other ABS. Nontraditional securitizations include nonprime RMBS and CDOs.
(a) Traditional vs. Nontraditional
(b) Traditional
Traditional vs. Non-Traditional Securitization
Traditional Securitization
100
Proceeds ($ billion)
Proceeds ($ billion)
400
300
200
100
80
60
40
20
2008q4
2007q4
2006q4
2005q4
2004q4
2003q4
CMBS
Consumer ABS
Non-traditional
(c) Nontraditional
Prime RMBS
Other ABS
(d) CDO
Collateralized Debt Obligations
Nontraditional Securitization
100
Proceeds ($ billion)
300
200
100
80
60
40
20
CBO
CDO
42
CLO
ABS CDO
2008q4
2007q4
2006q4
2005q4
2004q4
2003q4
2002q4
2001q4
2000q4
1999q4
2008q4
2007q4
2006q4
2005q4
2004q4
2003q4
2001q4
2000q4
1999q4
2002q4
Nonprime RMBS
1998q4
0
0
1998q4
Proceeds ($ billion)
2002q4
2001q4
2000q4
1998q4
2008q4
2006q4
2004q4
2002q4
2000q4
1998q4
1996q4
1994q4
1992q4
1990q4
1988q4
Traditional
1999q4
0
0
Figure 2
Credit Spreads on Securitizations and Corporate Bonds, 2003-2007
This figure plots the credit spreads on newly issued securitizations versus the spreads on corporate bonds. Each quarter we compute the value-weighted average credit spread on securitizations,
based on the underlying collateral. We compute this average for traditional consumer asset-backed
securities backed by credit card loans, automotive loans, student loans, and other personal loans
(“Consumer ABS”); private residential mortgage backed securities (RMBS) backed by subprime and
Alt-A mortgages (“Non-Prime RMBS”); commercial mortgage backed securities (CMBS); and collateralized debt obligations (“CDOs”). We compute value-weighted averages of new-issue spreads,
restricting attention to the spreads on floating rate notes indexed to 1-, 3-, and 6-month LIBOR
to avoid benchmarking issues. In Panel A, we plot the credit spreads on newly issued AAA-rated
securitizations. For reference, we plot the average secondary spreads over LIBOR (based on interest rate swaps) on 3-year AAA-rated and BBB-rated corporate bonds from Barclays. In Panel
B, we plot the credit spreads on newly issued BBB-rated securitizations alongside the spreads on
BBB-rated, BB-rated and B-rated corporate bonds.
(a) AAA
(b) BBB
500
100
450
400
80
350
bps
bps
300
60
250
200
40
150
100
20
50
0
0
2003q2
2003q2
2004q2
2003q4
2005q2
2004q4
AAA Consumer ABS
AAA Non-Prime RMBS
AAA Corporate
2006q2
2005q4
2004q2
2003q4
2007q2
2006q4
2007q4
AAA CMBS
AAA CDOs
BBB Corporate
43
2005q2
2004q4
BBB Consumer ABS
BBB Non-Prime RMBS
BBB Corporate
B Corporate
2006q2
2005q4
2007q2
2006q4
BBB CMBS
BBB CDOs
BB Corporate
2007q4
Figure 3
Ex-Ante Spreads versus Realized Loss Rates
This figure shows realized loss rates versus average spreads during the boom, 2003–2007. Realized
loss rates are from Moody’s (2012). Corporate spreads are from Bloomberg. For securitizations, we
compute value-weighted averages of new-issue spreads for securities in our eMAXX data, restricting
attention to the spreads on floating rate notes indexed to 1-, 3-, and 6-month LIBOR to avoid
benchmarking issues.
(a)
(b)
25
AAA Securitization vs. Corporate Spreads
8
M
BS
R
e
BB
B
N
on
-
pr
im
70
60
O
50
BB
B
C
D
40
or
at
e
or
p
B
C
or B
at B
e B
or
p
C
BB
C
B
BB
AA
AAA
C
A Coorp
C rp o
or o ra
po ra te
ra te
te
BB
B
C
or
po
ra
te
on
su
m
C
er
A
M
BS
BS
30
20
10
0
M
BS
R
e
pr
im
1
1.5
Average Spread in Boom
2
2.5
44
or
at
e
on
N
A
.5
1
Average Spread in Boom
BB
C
or
p
AA
0
Fitted values
BBB Securitization vs. Corporate Spreads
Cumulative 5 yr Loss
AA
A
6
(c)
.5
AA
AAA C
A orp
C o
on ra
t
AA sume
er
A
C
A
M
BS BS
BB
B
C
or
po
ra
te
5
4
Average Spread in Boom
0
2
Cumulative 5 yr Loss
0
15
20
or
p
C
B
or
at
e
0
10
or
at
e
Cumulative 5 yr Loss
30
20
10
AAA
AAA
BCBCoCoo
Brprporpoo
C rarara
or tetete
po
ra
te
BB
C
or
p
0
Cumulative 5 yr Loss
C
C
C
D
C
O
Co
r
po
ra
40
te
Corporate Spreads
1.5
Figure 4
Rating Composition of Mutual Fund Portfolios
For each mutual fund-quarter observation we calculate the fraction of the portfolio invested in
tranches with di↵erent ratings. We then calculate the average across funds. We do this separately
for di↵erent types of securitizations: consumer ABS, CDO, CMBS, and nonprime RMBS. We use
at-issuance ratings from Moody’s and restrict the sample to deals rated by Moody’s. NR means
that the overall securitization deal was rated by Moody’s but the specific tranche was not.
(a) Consumer ABS
(b) CDO
.8
.5
.4
.6
.3
.4
.2
.2
.1
0
0
2003q4
2004q4
2005q4
Aaa
Ba
2006q4
2007q4
Aa
B
2008q4
A
Caa
2009q4
2010q4
2003q4
Baa
NR
2004q4
2005q4
Aaa
Ba
(c) CMBS
2006q4
Aa
B
2007q4
2008q4
A
Caa
2009q4
2010q4
Baa
NR
(d) Nonprime RMBS
1
.8
.8
.6
.6
.4
.4
.2
.2
0
0
2003q4
2004q4
Aaa
Ba
2005q4
2006q4
Aa
B
2007q4
A
Caa
2008q4
2009q4
2010q4
Baa
NR
2003q4
2004q4
Aaa
Ba
45
2005q4
2006q4
Aa
B
2007q4
A
Caa
2008q4
2009q4
Baa
NR
2010q4
Figure 5
Rating Composition of Insurance Company Portfolios
For each insurance firm-quarter observation we calculate the fraction of the portfolio invested in
tranches with di↵erent ratings. We then calculate the average across insurance firms. We do this
separately for di↵erent types of securitizations: consumer ABS, CDO, CMBS, and nonprime RMBS.
We use at-issuance ratings from Moody’s and restrict the sample to deals rated by Moody’s. NR
means that the overall securitization deal was rated by Moody’s but the specific tranche was not.
(a) Consumer ABS
(b) CDO
1
.6
.8
.4
.6
.4
.2
.2
0
0
2003q4
2004q4
2005q4
Aaa
Ba
2006q4
2007q4
Aa
B
2008q4
A
Caa
2009q4
2010q4
2003q4
Baa
NR
2004q4
2005q4
Aaa
Ba
(c) CMBS
2006q4
Aa
B
2007q4
2008q4
A
Caa
2009q4
2010q4
Baa
NR
(d) Nonprime RMBS
1
1
.8
.8
.6
.6
.4
.4
.2
.2
0
0
2003q4
2004q4
Aaa
Ba
2005q4
2006q4
Aa
B
2007q4
A
Caa
2008q4
2009q4
2010q4
Baa
NR
2003q4
2004q4
Aaa
Ba
46
2005q4
2006q4
Aa
B
2007q4
A
Caa
2008q4
2009q4
Baa
NR
2010q4
Figure 6
Nontraditional Share of Variable Annuities versus Regular Mutual Funds
The sample consists of matched pairs of variable annuities and regular mutual funds that have the
same investment objectives and are managed by the same portfolio managers. We form the sample
of matched pairs as of 2003Q4.
.07
160
140
.05
120
.04
100
.03
.02
80
2003q3 2004q3 2005q3 2006q3 2007q3 2008q3 2009q3 2010q3
Variable Annuity Funds
# Fund Pairs
47
Regular Mutual Funds
# Fund Pairs
Nontraditional share
.06
Figure 7
Mutual Fund Manager Experience and Nontraditional Share
This figure shows the average nontraditional share of mutual funds managed by experienced versus
inexperienced portfolio managers. We split mutual funds into two groups based on the median
value of manager experience in Panel A and of manager age in Panel B. Manager characteristics
are measured as of 2004Q4. For team-managed funds we take the average of individual managers.
(b) Age
.08
.08
.07
.07
Nontraditional share
Nontraditional share
(a) Experience
.06
.05
.04
.06
.05
.04
.03
.03
.02
.02
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Inexperienced
Experienced
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Young
48
Old
Figure 8
Experience and Nontraditional Share in Individually- versus Team-Managed Funds
This figure shows the average nontraditional share of mutual funds managed by experienced versus
inexperienced individual managers and teams. We separately split individually- and team-managed
funds into experienced and inexperienced ones based on their median value of experience. Manager characteristics are measured as of 2004Q4. For team-managed funds we take the average of
individual managers.
.1
Nontraditional share
.08
.06
.04
.02
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Inexperienced individual
Inexperienced team
49
Experienced individual
Experienced team
Figure 9
Insurance Company Characteristics and Nontraditional Share
This figure shows the average nontraditional share of insurance companies split based on di↵erent
characteristics. Rating is the 2002 financial strength rating from AM Best.
(a) Type
(b) Rating
0.080
0.080
0.070
0.060
0.060
0.040
0.050
0.020
0.040
0.000
0.030
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Life
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
P&C
Weak rating
(c) Capital
Strong rating
(d) Mutual
0.070
0.060
0.060
0.050
0.050
0.040
0.040
0.030
0.030
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Weak capital
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Strong capital
Non mutual
(e) Public
Mutual
(f ) External
0.060
.06
0.055
.05
0.050
0.045
.04
0.040
.03
0.035
2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4
Private
Public
2003q3
2004q3
2005q3
2006q3
Internal
50
2007q3
2008q3
External
2009q3
2010q3
Figure 10
Insurance Company Portfolio Turnover
Turnover is the absolute value of transactions, buys and sells, normalized by average holdings during
quarters t 1 and t. Transactions in private securitizations and in GSE MBS are from eMAXX.
Transactions in corporate bonds are from Mergent FISD.
(a) ABS
(b) CDO
.3
.3
.25
.2
.2
.15
.1
.1
.05
0
2003q4
2004q4
2005q4
Agencies
2006q4
2007q4
2008q4
Corporates
2009q4
2010q4
2003q4
ABS
2004q4
2005q4
Agencies
(c) CMBS
2006q4
2007q4
2008q4
Corporates
2009q4
2010q4
ABS
(d) Nonprime RMBS
.3
.3
.25
.25
.2
.2
.15
.15
.1
.1
.05
.05
2003q4
2004q4
2005q4
Agencies
2006q4
2007q4
Corporates
2008q4
2009q4
2010q4
ABS
2003q4
2004q4
2005q4
Agencies
51
2006q4
2007q4
Corporates
2008q4
2009q4
ABS
2010q4
Figure 11
Mutual Fund Portfolio Turnover
Turnover is the absolute value of transactions, buys and sells, normalized by average holdings during
quarters t 1 and t. Sales and purchases are calculated as the change in par holdings adjusted for
the change in security’s outstanding amount using outstanding factors from Bloomberg. Interestand principal-only tranches are excluded using the following screens: a) zero original balance from
any of the three rating agencies; b) tranche name; c) fund’s holdings of a security are greater than
150% of outstanding balance; d) security weight in the fund’s portfolio is greater than 10%; e)
outstanding factor is less than 0.01 or greater than 2; f) fund’s holdings of the security are greater
than $100 million. The sample of funds consists of bond and hybrid funds matched to CRSP.
.8
.6
.4
.2
0
2003q4
2004q4
2005q4
2006q4
GSE MBS
CDO
Nonprime RMBS
52
2007q4
2008q4
2009q4
Consumer ABS
CMBS
2010q4
Table 1
Outstanding asset-backed securities and holdings of insurers and mutual funds by underlying collateral type, 2003-2010
This table reports estimates of the outstanding balance of US asset-backed securities by underlying collateral type. The sources
for estimated outstanding are listed in the footnotes. For each collateral type, we compute aggregate par holdings by insurers and mutual
funds using our eMaxx holdings data. For the market, % refers to the share of a given collateral type in the aggregate outstanding
amount of all private credit securities. For insurance and mutual funds, % refers to the shares of a given collateral type in the aggregate
portfolios of insurance firms and mutual funds.
All private credit securitiesa
53
2003Q2
2003Q4
2004Q2
2004Q4
2005Q2
2005Q4
2006Q2
2006Q4
2007Q2
2007Q4
2008Q2
2008Q4
2009Q2
2009Q4
2010Q2
2010Q4
Market
$
%
6,776 100.0%
7,178 100.0%
7,524 100.0%
8,071 100.0%
8,541 100.0%
8,852 100.0%
9,415 100.0%
10,103 100.0%
10,879 100.0%
11,577 100.0%
11,698 100.0%
11,170 100.0%
11,676 100.0%
11,654 100.0%
11,432 100.0%
11,816 100.0%
Insurance
$
%
1,755 25.9%
1,829 25.5%
1,927 25.6%
2,002 24.8%
2,060 24.1%
2,088 23.6%
2,127 22.6%
2,097 20.8%
2,132 19.6%
2,145 18.5%
2,143 18.3%
2,085 18.7%
2,146 18.4%
2,226 19.1%
2,288 20.0%
2,353 19.9%
Mutual
Funds
$
%
504
7.4%
531
7.4%
560
7.4%
594
7.4%
631
7.4%
660
7.5%
706
7.5%
764
7.6%
837
7.7%
886
7.7%
958
8.2%
956
8.6%
1,035
8.9%
1,121
9.6%
1,181 10.3%
1,243 10.5%
Traditional consumer ABSb
Mutual
Market
Insurance
Funds
$
%
$
%
$
%
701 10.3%
93 5.3% 22 4.4%
721 10.0% 100 5.5% 20 3.8%
726
9.7% 101 5.3% 25 4.4%
729
9.0%
98 4.9% 24 4.1%
762
8.9%
90 4.4% 24 3.8%
781
8.8%
89 4.3% 28 4.3%
794
8.4%
84 4.0% 23 3.3%
836
8.3%
76 3.6% 24 3.1%
870
8.0%
64 3.0% 27 3.2%
894
7.7%
68 3.2% 26 2.9%
894
7.6%
85 4.0% 34 3.6%
835
7.5%
75 3.6% 27 2.8%
808
6.9%
68 3.2% 26 2.5%
798
6.8%
70 3.1% 23 2.1%
744
6.5%
75 3.3% 32 2.7%
690
5.8%
68 2.9% 28 2.3%
CMBSc
Market
$
%
313 4.6%
343 4.8%
368 4.9%
394 4.9%
439 5.1%
506 5.7%
553 5.9%
614 6.1%
697 6.4%
758 6.5%
741 6.3%
718 6.4%
703 6.0%
671 5.8%
647 5.7%
617 5.2%
Insurance
$
%
108
6.1%
119
6.5%
130
6.7%
145
7.2%
153
7.4%
168
8.1%
185
8.7%
189
9.0%
195
9.1%
216 10.1%
222 10.4%
223 10.7%
221 10.3%
227 10.2%
220
9.6%
208
8.8%
Mutual
Funds
$
%
10 2.0%
11 2.0%
11 2.0%
13 2.1%
15 2.3%
21 3.2%
19 2.7%
24 3.2%
38 4.5%
41 4.6%
48 5.0%
50 5.3%
48 4.7%
52 4.6%
58 4.9%
59 4.8%
a
Data on the outstanding quantity of all private credit securities, which include corporate bonds, foreign bonds, and privately issued ABS is from
table L.212 ‘Corporate and Foreign Bonds’ of the Federal Reserve’s Financial Accounts of the United States (i.e. the Z1 release which was formerly
known as the Flow of Funds Accounts). Data on the holding of insurers (which includes both P&C and life insurers) and mutual funds (which excludes
money market mutual funds) is also from Table L.212.
b
Data on the outstanding quantity of traditional consumer ABS, which consists of ABS collateralized by credit card debt, automotive loans, student
loans, manufactured housing and equipment loans, is from SIFMA at
http://www.sifma.org/uploadedFiles/Research/Statistics/StatisticsFiles/SF-US-ABS-SIFMA.xls.
c
Data on the outstanding quantity of CMBS, which consists of MBS collateralized by commercial mortgages and multifamily residential mortgages,
is from tables L.219 ‘Multifamily Residential Mortgages’ and Table L.220 ‘Commercial Mortgages’ of the Federal Reserve’s Financial Accounts of the
United States.
54
Insurance
$
%
29
1.6%
22
1.2%
29
1.5%
51
2.5%
55
2.7%
62
3.0%
69
3.2%
74
3.5%
73
3.4%
84
3.9%
83
3.9%
82
3.9%
74
3.4%
67
3.0%
62
2.7%
54
2.3%
Mutual
Funds
$
%
6
1.3%
5
0.9%
10
1.7%
10
1.7%
11
1.7%
13
2.0%
12
1.7%
15
1.9%
19
2.2%
25
2.8%
23
2.4%
22
2.3%
18
1.8%
16
1.5%
20
1.7%
22
1.8%
Market
$
%
563
8.3%
601
8.4%
765
10.2%
929
11.5%
1,139
13.3%
1,349
15.2%
1,579
16.8%
1,808
17.9%
1,852
17.0%
1,895
16.4%
1,729
14.8%
1,563
14.0%
1,437
12.3%
1,311
11.2%
1,221
10.7%
1,131
9.6%
Insurance
$
%
73
4.2%
56
3.1%
65
3.4%
83
4.2%
96
4.7%
109
5.2%
117
5.5%
122
5.8%
115
5.4%
130
6.1%
138
6.4%
136
6.5%
126
5.9%
125
5.6%
119
5.2%
107
4.6%
Subprime RMBSb
Mutual
Funds
$
%
8
1.6%
8
1.5%
10
1.8%
14
2.4%
16
2.5%
18
2.7%
14
2.0%
16
2.1%
19
2.3%
17
1.9%
18
1.9%
16
1.7%
13
1.3%
11
1.0%
13
1.1%
14
1.1%
Market
$
%
279
4.1%
302
4.2%
347
4.6%
391
4.8%
454
5.3%
517
5.8%
627
6.7%
795
7.9%
961
8.8%
1,005
8.7%
1,004
8.6%
1,018
9.1%
987
8.5%
936
8.0%
872
7.6%
838
7.1%
Insurance
$
%
21
1.2%
20
1.1%
21
1.1%
23
1.1%
21
1.0%
24
1.2%
24
1.1%
23
1.1%
21
1.0%
21
1.0%
27
1.3%
27
1.3%
26
1.2%
26
1.2%
25
1.1%
24
1.0%
CDOsc
Mutual
Funds
$
%
4
0.8%
5
0.9%
6
1.0%
5
0.9%
3
0.5%
2
0.4%
2
0.2%
2
0.2%
2
0.3%
2
0.3%
4
0.4%
2
0.2%
2
0.2%
1
0.1%
2
0.1%
2
0.1%
Data on the outstanding quantity of CBOs, CDOs, and CLOs is from SIFMA at
http://www.sifma.org/uploadedFiles/Research/Statistics/StatisticsFiles/SF-Global-CDO-SIFMA.xls.
SIFMA provides data on global CBOs, CDOs, and CLOs outstandings. Based on SIFMA’s issuance data by currency, we assume that 75% of
outstanding CDOs and CLOs are USD-denominated.
c
Data on the outstanding quantity of subprime RMBS, a subset of all private RMBS, is from SIFMA at
http://www.sifma.org/uploadedFiles/Research/Statistics/StatisticsFiles/SF-US-ABS-SIFMA.xls.
Specifically, we use the series for ‘home equity ABS’—market-speak for subprime RMBS.
b
Data on the outstanding quantity of private RMBS, which consists of private MBS collateralized by first lien single-family residential mortgages, is
from table L.218 ‘Home Mortgages’ of Federal Reserve’s Financial Accounts of the United States.
a
2003Q2
2003Q4
2004Q2
2004Q4
2005Q2
2005Q4
2006Q2
2006Q4
2007Q2
2007Q4
2008Q2
2008Q4
2009Q2
2009Q4
2010Q2
2010Q4
Market
$
%
297
4.4%
276
3.8%
314
4.2%
351
4.3%
388
4.5%
425
4.8%
462
4.9%
499
4.9%
525
4.8%
552
4.8%
518
4.4%
484
4.3%
436
3.7%
389
3.3%
350
3.1%
311
2.6%
All Private RMBSa
This table reports estimates of the outstanding balance of US asset-backed securities by underlying collateral type. The sources
for estimated outstanding are listed in the footnotes. For each collateral type, we compute aggregate par holdings by insurers and mutual
funds using our eMaxx holdings data. For the market, % refers to the share of a given collateral type in the aggregate outstanding
amount of all private credit securities. For insurance and mutual funds, % refers to the shares of a given collateral type in the aggregate
portfolios of insurance firms and mutual funds.
Table 2
Outstanding asset-backed securities and holdings of insurers and mutual funds by underlying collateral type, 2003-2010
Table 3
Variance Decomposition of Mutual Funds’ Nontraditional Share
This table reports the R2 from the regressions of mutual funds’ nontraditional share on various
fixed e↵ects. The sample of funds consists of bond and hybrid mutual funds. The sample period
is 2003Q1–2010Q4.
Adjusted R
N
2
Date Objective
0.017
0.145
26550
26550
Objective ⇥
Date
0.175
26550
55
Family
Fund
All
5 Funds 10 Funds
0.623 0.312
0.231
0.116
26550 24331
18454
11762
Table 4
Performance-Flow Relationship in Variable Annuity versus Regular Mutual Funds
This table reports the results of the performance-flow regressions for variable annuity versus regular
mutual funds. The sample consists of all bond and hybrid funds in CRSP Mutual Fund Database.
The sample period is January 2009–June 2013. Fund flows are winsorized at the 1st and 99th
percentiles in columns 1–2 and at the 5th and 95th percentiles in columns 3–4. Objective-month
date fixed e↵ects are included in all specifications. Standard errors are adjusted for clustering by
fund in columns 2–4. *, **, and *** indicate statistical significance at 10%, 5%, and 1%.
Variable annuity
Rett
1⇥
Regular
Rett
1⇥
Variable annuity
Log(TNAt
Rett
1⇥
(1)
0.003⇤⇤⇤
(0.001)
0.192⇤⇤⇤
(0.025)
0.057⇤
(0.031)
(2)
0.003⇤⇤
(0.002)
0.192⇤⇤⇤
(0.034)
0.057
(0.036)
(3)
0.001
(0.001)
0.121⇤⇤⇤
(0.016)
0.057⇤⇤⇤
(0.016)
0.018⇤⇤⇤
(0.000)
323208
0.016
0.018⇤⇤⇤
(0.001)
323208
0.016
0.009⇤⇤⇤
(0.000)
323208
0.036
1)
Log(TNAt
Constant
N
Adjusted R2
1)
56
(4)
0.000
(0.001)
0.111⇤⇤⇤
(0.018)
0.044⇤⇤
(0.018)
0.001⇤⇤⇤
(0.000)
0.005⇤⇤⇤
(0.002)
0.012⇤⇤⇤
(0.001)
323208
0.038
Table 5
Summary Statistics: Mutual Fund Managers
This table reports summary statistics for mutual fund managers in our data. The unit of observation
is fund-quarter. The sample of funds consists of bond and hybrid mutual funds. The sample
period is 2003Q1–2010Q4. Fund manager characteristics are measured as of December 31, 2004.
Team managed
Number of fund managers
Age
Experience
N
23377
23377
15379
22832
Mean
0.7
2.7
45.2
8.5
57
Median
1.0
2.0
44.0
7.9
SD
0.5
2.5
7.1
4.5
Min
0.0
0.0
29.0
0.0
Max
1.0
21.0
71.0
29.1
Table 6
Nontraditional Share of Mutual Funds
This table reports the results of the regressions of mutual funds’ nontraditional share on fund and
portfolio manager characteristics. Experienced is a binary variable equal to 1 for funds above the
median of experience. Manager characteristics are measured as of 2004Q4. Retail share is the share
of retail share classes in fund assets. Family retail share is the share of retail share classes in fund
family assets. Family ETF, equity, and taxable shares are the shares of ETFs, equity funds, and
taxable bond funds in fund family assets. The omitted category is municipal funds. Objective fixed
e↵ects are included. Standard errors are adjusted for clustering by fund. *, **, and *** indicate
statistical significance at 10%, 5%, and 1%.
Experienced
Log(TNA)
Log(Family TNA)
Log(Fund age)
Retail share
Family retail share
Family ETF share
Family equity share
Family taxable share
Constant
N
Adjusted R2
2004
0.006⇤
(0.004)
0.002
(0.002)
0.002
(0.001)
0.002
(0.003)
0.007
(0.007)
0.003
(0.008)
0.005
(0.024)
0.003
(0.012)
0.031⇤⇤
(0.014)
0.006
(0.014)
3102
0.287
2005
0.009⇤⇤
(0.004)
0.003
(0.002)
0.001
(0.001)
0.010⇤⇤
(0.004)
0.019⇤⇤
(0.008)
0.019⇤
(0.010)
0.018
(0.034)
0.003
(0.027)
0.026
(0.028)
0.042
(0.032)
2969
0.303
58
2006
0.020⇤⇤⇤
(0.006)
0.005⇤
(0.003)
0.000
(0.002)
0.011⇤
(0.007)
0.020⇤
(0.010)
0.035⇤⇤
(0.014)
0.017
(0.035)
0.001
(0.047)
0.033
(0.048)
0.062
(0.054)
2915
0.272
2007
0.034⇤⇤⇤
(0.007)
0.005⇤
(0.003)
0.002
(0.002)
0.013⇤
(0.008)
0.010
(0.010)
0.006
(0.015)
0.049
(0.045)
0.019
(0.028)
0.087⇤⇤
(0.036)
0.055⇤
(0.031)
2935
0.185
2008
0.023⇤⇤⇤
(0.007)
0.001
(0.003)
0.001
(0.002)
0.003
(0.006)
0.009
(0.010)
0.018
(0.016)
0.068⇤⇤
(0.029)
0.019
(0.036)
0.097⇤⇤
(0.044)
0.024
(0.035)
2888
0.177
2009
0.001
(0.006)
0.003
(0.002)
0.001
(0.002)
0.006
(0.005)
0.009
(0.014)
0.021
(0.022)
0.096⇤⇤
(0.038)
0.026
(0.032)
0.054
(0.034)
0.023
(0.034)
2706
0.118
Table 7
Manager Experience and Performance-Flow Relationship
This table reports the results of monthly performance-flow regressions for bond and hybrid mutual
funds in our sample. The sample period is 2003–2010. Fund manager characteristics are measured
as of December 31, 2004. Fund flows are winsorized at the 1st and 99th percentiles. Objective-date
fixed e↵ects are included. Heteroskedasticity robust standard errors are reported. *, **, and ***
indicate statistical significance at 10%, 5%, and 1%.
Returnt
1
Log(TNA)t
Returnt
1
1⇥
Log(TNA)t
1⇥
Old
1
(1)
0.368⇤⇤⇤
(0.037)
0.002⇤⇤⇤
(0.000)
0.016⇤⇤⇤
(0.004)
Old
Returnt
(2)
0.379⇤⇤⇤
(0.043)
0.001⇤⇤⇤
(0.000)
0.015⇤⇤⇤
(0.004)
0.001⇤⇤
(0.000)
0.005
(0.017)
1⇥
Experienced
Team managed
Returnt
1⇥
Constant
N
Adjusted R2
(4)
0.362⇤⇤⇤
(0.040)
0.002⇤⇤⇤
(0.000)
0.016⇤⇤⇤
(0.005)
0.001⇤
(0.000)
0.010
(0.015)
Experienced
Returnt
(3)
0.367⇤⇤⇤
(0.037)
0.002⇤⇤⇤
(0.000)
0.016⇤⇤⇤
(0.004)
Team managed
0.012⇤⇤⇤
(0.001)
131465
0.056
59
0.010⇤⇤⇤
(0.001)
88402
0.052
0.012⇤⇤⇤
(0.001)
131465
0.056
0.001⇤⇤⇤
(0.000)
0.010
(0.015)
0.014⇤⇤⇤
(0.001)
134186
0.054
Table 8
Nonlinearity of Experience
This table reports the results of the regressions of mutual funds’ nontraditional share on portfolio
manager experience defined using di↵erent cuto↵s. Manager characteristics are measured as of
2004Q4. The sample period is 2007Q1–2007Q4. Same controls as in Table 6 are included but not
reported. Objective fixed e↵ects are included. Standard errors are adjusted for clustering by fund.
*, **, and *** indicate statistical significance at 10%, 5%, and 1%.
Experienced
N
Adjusted R2
2001
0.000
(0.011)
2935
0.164
Started managing mutual funds
2000
1999
1998
0.002
0.005
0.014⇤
(0.009)
(0.008)
(0.007)
2935
2935
2935
0.164
0.164
0.167
60
before January 1
1997
1996
⇤⇤⇤
0.032
0.023⇤⇤⇤
(0.007)
(0.007)
2935
2935
0.182
0.173
1995
0.027⇤⇤⇤
(0.006)
2935
0.176
Table 9
Teams, Experience, and Nontraditional Share
This table reports the results of the regressions of mutual funds’ nontraditional share on fund and
portfolio manager characteristics. The sample period is 2007Q1–2007Q4. Experienced is a binary
variable equal to 1 for funds above the median of experience. Manager characteristics are measured
as of 2006Q4. Retail share is the share of retail share classes in fund assets. Family retail share
is the share of retail share classes in fund family assets. Family ETF, equity, and taxable shares
are the shares of ETFs, equity funds, and taxable bond funds in fund family assets. The omitted
category is municipal funds. Objective fixed e↵ects are included. Standard errors are adjusted for
clustering by fund. *, **, and *** indicate statistical significance at 10%, 5%, and 1%.
Log(Experience)
(1)
0.004
(0.004)
Log(Manager age)
(2)
0.068
(0.041)
(3)
0.009
(0.010)
0.053
(0.050)
(4)
(5)
0.021⇤⇤⇤
(0.007)
Experienced
0.022⇤⇤
(0.011)
0.027⇤⇤⇤
(0.010)
Experience percentile
Team
Experienced ⇥ Team
Log(TNA)
Log(Family TNA)
Log(Fund age)
Retail share
Family retail share
Family ETF share
Family equity share
Family taxable share
Constant
N
Adjusted R2
(6)
0.006⇤⇤
(0.003)
0.003
(0.003)
0.012⇤
(0.007)
0.005
(0.010)
0.007
(0.015)
0.081⇤
(0.046)
0.022
(0.038)
0.045
(0.039)
0.087⇤
(0.050)
3224
0.163
0.006
(0.004)
0.005
(0.004)
0.017
(0.010)
0.004
(0.013)
0.004
(0.020)
0.944
(0.722)
0.029
(0.054)
0.062
(0.051)
0.365⇤⇤
(0.183)
2187
0.181
61
0.007
(0.004)
0.004
(0.004)
0.016
(0.010)
0.004
(0.013)
0.004
(0.020)
0.836
(0.714)
0.026
(0.054)
0.066
(0.052)
0.322
(0.203)
2187
0.182
0.007⇤⇤
(0.003)
0.003
(0.003)
0.012⇤
(0.007)
0.003
(0.010)
0.005
(0.015)
0.086⇤
(0.047)
0.021
(0.038)
0.047
(0.039)
0.084⇤
(0.049)
3224
0.170
0.007⇤⇤
(0.003)
0.003
(0.003)
0.012⇤
(0.007)
0.004
(0.010)
0.005
(0.015)
0.083⇤
(0.047)
0.020
(0.038)
0.048
(0.038)
0.085⇤
(0.049)
3224
0.166
0.027⇤⇤
(0.013)
0.005
(0.015)
0.007⇤⇤
(0.003)
0.002
(0.003)
0.011
(0.007)
0.004
(0.010)
0.002
(0.015)
0.091⇤
(0.047)
0.014
(0.037)
0.055
(0.038)
0.050
(0.050)
3224
0.182
Table 10
Variance Decomposition of Insurance Companies’ Nontraditional Share
This table reports the R2 from the regressions of insurance companies’ nontraditional share on date,
firm, and financial group fixed e↵ects. Financial group fixed e↵ect refers to the ultimate parent
according to AM Best.
Adjusted R
N
2
Date
0.002
77272
Firm
0.582
77272
62
Group
0.458
77272
Table 11
Nontraditional Share and Capital Ratio
This table reports the results of the regressions of insurance companies’ nontraditional share. The
change in nontraditional share is calculated relative to 2004. Size is the log of the par value of
the bond portfolio. Standard errors are adjusted for clustering by firm. *, **, and *** indicate
statistical significance at 10%, 5%, and 1%.
2004
Capital/Assets2003
Size
Constant
N
Adjusted R2
Capital/Assets2003
Size
Constant
N
Adjusted R2
2005
2006
2007
Panel A: Nontraditional Share
0.071⇤⇤⇤
0.098⇤⇤⇤
0.119⇤⇤⇤
0.122⇤⇤⇤
(0.017)
(0.019)
(0.021)
(0.025)
0.002
0.003⇤
0.003
0.004⇤
(0.002)
(0.002)
(0.002)
(0.002)
0.024
0.020
0.033
0.027
(0.019)
(0.018)
(0.020)
(0.023)
967
891
834
742
0.144
0.217
0.205
0.210
Panel B: Change in Nontraditional Share
0.025⇤⇤⇤
0.039⇤⇤⇤
0.045⇤⇤⇤
(0.009)
(0.015)
(0.016)
0.003⇤⇤
0.003
0.004⇤
(0.001)
(0.002)
(0.002)
0.018
0.009
0.016
(0.013)
(0.019)
(0.021)
886
824
733
0.130
0.082
0.110
63
2008
2009
0.117⇤⇤⇤
(0.027)
0.005⇤⇤
(0.002)
0.021
(0.023)
732
0.189
0.099⇤⇤⇤
(0.031)
0.003
(0.002)
0.033
(0.025)
732
0.111
0.037⇤
(0.020)
0.005⇤⇤
(0.002)
0.022
(0.021)
721
0.086
0.020
(0.026)
0.003⇤
(0.002)
0.011
(0.021)
712
0.027
Table 12
Nontraditional Share and Financial Strength Rating
This table reports the results of the regressions of insurance companies’ nontraditional share. The
change in nontraditional share is calculated relative to 2004. Strong rating is a binary variable
equal to one for firms whose 2003 AM Best financial strength rating is above the median. Size is
the log of the par value of the bond portfolio. Standard errors are adjusted for clustering by firm.
*, **, and *** indicate statistical significance at 10%, 5%, and 1%.
2004
Strong rating
Size
Constant
N
Adjusted R2
Strong rating
Size
Constant
N
Adjusted R2
2005
2006
2007
Panel A: Nontraditional Share
0.003
0.012
0.018⇤
0.023⇤⇤
(0.008)
(0.009)
(0.010)
(0.011)
⇤⇤
⇤⇤⇤
⇤⇤⇤
0.004
0.007
0.008
0.009⇤⇤⇤
(0.002)
(0.002)
(0.002)
(0.003)
0.014
0.032⇤
0.030
0.038
(0.016)
(0.019)
(0.021)
(0.023)
831
768
726
652
0.064
0.131
0.127
0.160
Panel B: Change in Nontraditional Share
0.013⇤⇤
0.019⇤⇤
0.025⇤⇤⇤
(0.006)
(0.008)
(0.008)
⇤⇤⇤
⇤⇤
0.004
0.005
0.006⇤⇤⇤
(0.002)
(0.002)
(0.002)
0.030⇤⇤
0.026
0.035⇤
(0.013)
(0.017)
(0.019)
766
722
650
0.231
0.160
0.237
64
2008
2009
0.027⇤⇤
(0.012)
0.010⇤⇤⇤
(0.003)
0.041⇤
(0.024)
639
0.173
0.021⇤
(0.012)
0.008⇤⇤⇤
(0.003)
0.026
(0.025)
633
0.101
0.029⇤⇤⇤
(0.009)
0.007⇤⇤⇤
(0.002)
0.038⇤⇤
(0.019)
635
0.230
0.023⇤⇤
(0.010)
0.005⇤⇤
(0.002)
0.023
(0.021)
621
0.117
Table 13
Nontraditional Share and Mutual Organization Form
This table reports the results of the regressions of insurance companies’ nontraditional share. The
change in nontraditional share is calculated relative to 2004. Mutual insurance firms are owned by
policyholders. Size is the log of the par value of the bond portfolio. Standard errors are adjusted
for clustering by firm. *, **, and *** indicate statistical significance at 10%, 5%, and 1%.
2004
Mutual
Size
Constant
N
Adjusted R2
Mutual
Size
Constant
N
Adjusted R2
2005
2006
2007
Panel A: Nontraditional Share
0.001
0.003
0.004
0.010
(0.011)
(0.017)
(0.020)
(0.019)
0.005⇤⇤⇤
0.007⇤⇤⇤
0.007⇤⇤⇤
0.008⇤⇤⇤
(0.002)
(0.002)
(0.002)
(0.002)
0.016
0.034⇤
0.033
0.040⇤
(0.016)
(0.018)
(0.021)
(0.021)
1045
1109
1093
997
0.070
0.106
0.088
0.105
Panel B: Change in Nontraditional Share
0.001
0.000
0.005
(0.009)
(0.012)
(0.012)
0.004⇤⇤
0.004⇤⇤
0.006⇤⇤
(0.002)
(0.002)
(0.002)
0.031⇤⇤
0.032
0.042⇤
(0.015)
(0.021)
(0.022)
954
880
778
0.094
0.055
0.084
65
2008
2009
0.009
(0.021)
0.009⇤⇤⇤
(0.002)
0.042⇤⇤
(0.021)
1117
0.103
0.003
(0.026)
0.007⇤⇤⇤
(0.002)
0.026
(0.023)
1251
0.058
0.005
(0.014)
0.006⇤⇤⇤
(0.002)
0.043⇤⇤
(0.021)
766
0.073
0.000
(0.019)
0.004⇤
(0.002)
0.023
(0.023)
757
0.024
Table 14
Nontraditional Share and External Managers
This table reports the results of the regressions of insurance companies’ nontraditional share. The
change in nontraditional share is calculated relative to 2004. External manager is a binary variable
equal to one for firms whose bond portfolios are managed by unaffiliated external managers. An
insurance company and an asset management firm are considered to be unaffiliated if they have
di↵erent ultimate parents in Capital IQ. Size is the log of the par value of the bond portfolio.
Standard errors are adjusted for clustering by firm. *, **, and *** indicate statistical significance
at 10%, 5%, and 1%.
2004
External manager
Size
Constant
N
Adjusted R2
External manager
Size
Constant
N
Adjusted R2
2005
2006
2007
Panel A: Nontraditional Share
0.007
0.005
0.010
0.008
(0.005)
(0.007)
(0.008)
(0.009)
0.002
0.006⇤⇤⇤
0.007⇤⇤⇤
0.007⇤⇤⇤
(0.001)
(0.002)
(0.002)
(0.002)
0.001
0.060⇤⇤⇤
0.063⇤⇤
0.062⇤⇤
(0.021)
(0.023)
(0.032)
(0.027)
2300
2265
2203
2088
0.021
0.094
0.085
0.091
Panel B: Change in Nontraditional Share
0.004
0.007
0.005
(0.005)
(0.007)
(0.007)
0.003⇤⇤⇤
0.005⇤⇤
0.005⇤⇤
(0.001)
(0.002)
(0.002)
0.044⇤⇤
0.055⇤
0.054⇤
(0.020)
(0.032)
(0.029)
2100
2007
1858
0.086
0.069
0.069
66
2008
2009
0.009
(0.009)
0.008⇤⇤⇤
(0.002)
0.066⇤⇤
(0.028)
2165
0.083
0.004
(0.010)
0.007⇤⇤⇤
(0.002)
0.052⇤
(0.030)
2219
0.043
0.008
(0.008)
0.005⇤⇤
(0.002)
0.054
(0.033)
1855
0.061
0.004
(0.009)
0.004⇤
(0.002)
0.041
(0.034)
1847
0.025
Table 15
Cross-Sectional Determinants of Insurance Company Transactions
This table reports the results of the regressions of net purchases of nontraditional securitizations and
of GSE MBS on various insurance company characteristics. Net purchases are scaled by the lagged
par value of the insurance firm’s fixed income portfolio. Crisis is 2007Q3–2009Q2. External manager
is a binary variable equal to one for firms whose bond portfolios are managed by unaffiliated external
managers. An insurance company and an asset management firm are considered to be unaffiliated
if they have di↵erent ultimate parents in Capital IQ. Strong rating is a binary variable equal to
one for firms whose 2003 AM Best financial strength rating is above the median. P&C is a binary
variable equal to one for property and casualty insurance. The omitted caterogy is life insurance.
Date fixed e↵ects are included. Standard errors are adjusted for clustering by firm. *, **, and ***
indicate statistical significance at 10%, 5%, and 1%.
Size
Size ⇥ Crisis
External
External ⇥ Crisis
Strong capital
Strong capital ⇥ Crisis
Strong rating
Strong rating ⇥ Crisis
P&C
P&C ⇥ Crisis
Constant
N
Adjusted R2
Net purchases of
Nontraditional securitizations
GSE MBS
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.020⇤⇤ 0.025⇤⇤⇤ 0.025⇤⇤⇤ 0.016⇤ 0.426⇤⇤⇤ 0.398⇤⇤⇤ 0.406⇤⇤⇤ 0.424⇤⇤⇤
(0.008) (0.008) (0.009) (0.009) (0.034) (0.032) (0.035) (0.036)
0.017
0.027⇤⇤ 0.027⇤⇤ 0.020
0.183⇤⇤⇤ 0.123⇤⇤ 0.169⇤⇤⇤ 0.192⇤⇤⇤
(0.014) (0.013) (0.014) (0.016) (0.067) (0.060) (0.062) (0.068)
0.025
0.281⇤
(0.039)
(0.158)
0.142
0.236
(0.088)
(0.277)
0.055
0.153
(0.036)
(0.134)
0.025
0.058
(0.074)
(0.222)
0.036
0.178
(0.042)
(0.155)
0.127⇤
0.324
(0.068)
(0.269)
0.028
0.624⇤⇤⇤
(0.041)
(0.151)
0.010
0.093
(0.084)
(0.283)
0.110⇤⇤ 0.013
0.043
0.138⇤⇤ 5.107⇤⇤⇤ 4.778⇤⇤⇤ 5.005⇤⇤⇤ 4.973⇤⇤⇤
(0.052) (0.042) (0.046) (0.061) (0.287) (0.227) (0.237) (0.280)
41956 39467 36063 44514 41956 39467 36063 44514
0.005
0.007
0.006
0.006
0.029
0.029
0.031
0.030
67
Table 16
Cross-Sectional Determinants of Mutual Fund Transactions
This table reports the results of the regressions of net purchases of nontraditional securitizations and
of GSE MBS on mutual fund flows. Net purchases of nontraditional securitizations are calculated
as the change in par holdings adjusted for the change in security’s outstanding amount using
outstanding factors from Bloomberg. Net purchases of GSE MBS are calculated similarly except
that we use aggregate instead of security-specific prepayment information. Specifically we calculate
aggregate prepayment speed as gross issuance of Fannie and Freddie MBS minus the change in
outstanding amount. Net purchases are scaled by the lagged par value of the mutual fund’s bond
portfolio. Crisis is 2007Q3–2009Q2. Fund flows are scaled by lagged total net assets, and are
winsorized at the 1st and 99th percentiles. Inflows are equal to max(0, Fund flows). Outflows are
equal to max(0, Fund flows). Size is the log of the par value of the bond portfolio. Objective-date
fixed e↵ects are included. Standard errors are adjusted for clustering by fund. *, **, and ***
indicate statistical significance at 10%, 5%, and 1%.
Inflows
Inflows ⇥ Crisis
Outflows
Outflows ⇥ Crisis
Size
Size ⇥ Crisis
Nontraditional share
Nontraditional share ⇥ Crisis
Constant
N
Adjusted R2
Net purchases of
Nontraditional
securitizations
GSE MBS
(1)
(2)
(3)
(4)
0.216⇤⇤⇤
0.187⇤⇤⇤
0.423⇤⇤⇤
0.390⇤⇤⇤
(0.048)
(0.046)
(0.054)
(0.054)
0.076
0.040
0.235⇤⇤
0.245⇤⇤⇤
(0.072)
(0.071)
(0.094)
(0.094)
⇤⇤⇤
0.032
0.105
0.194
0.263⇤⇤⇤
(0.071)
(0.069)
(0.063)
(0.063)
⇤
0.056
0.040
0.283
0.273⇤
(0.097)
(0.096)
(0.150)
(0.159)
0.005⇤⇤⇤
0.006⇤⇤⇤
(0.001)
(0.001)
0.004⇤⇤⇤
0.001
(0.001)
(0.001)
0.176⇤⇤⇤
0.024⇤⇤
(0.026)
(0.011)
⇤⇤⇤
0.104
0.005
(0.028)
(0.024)
0.014⇤⇤⇤
0.056⇤⇤⇤
0.022⇤⇤⇤
0.088⇤⇤⇤
(0.001)
(0.007)
(0.001)
(0.008)
20076
20076
23537
23537
0.047
0.083
0.072
0.078
68
Download